Fixed performance issue with IIR filter
Passing samples with zero value (0.0) causes very slow processing of IIR filter on my test machine 'nuedel'. I don't know why. To solve this, I add a very small number to each input sample.
This commit is contained in:
@@ -92,7 +92,6 @@ int fm_mod_init(fm_mod_t *mod, double samplerate, double offset, double amplitud
|
||||
/* generate ramp up with ramp_length */
|
||||
for (i = 0; i < mod->ramp_length; i++)
|
||||
mod->ramp_tab[i] = 0.5 - cos(M_PI * i / mod->ramp_length) / 2.0;
|
||||
mod->ramp_tab[0] = mod->ramp_tab[1] / 2.0; /* never be 0 */
|
||||
|
||||
return 0;
|
||||
}
|
||||
@@ -199,10 +198,6 @@ again:
|
||||
dev = offset + *frequency++;
|
||||
power++;
|
||||
length--;
|
||||
/* somehow we need to have some value, otherwise IIR filter will be very slow!
|
||||
* we still continue with a carrier, but it has very low amplitude.
|
||||
* the low amplitude is set in ramp_tab[0]
|
||||
*/
|
||||
if (fast_math) {
|
||||
phase += 65536.0 * dev / rate;
|
||||
if (phase < 0.0)
|
||||
|
Reference in New Issue
Block a user