Hi Yoann, As James pointed out, Faust has strict semantics. This means that all signals are always computed (https://en.wikipedia.org/wiki/Strict_function) and that there is no non-strict 'if' in Faust.
It turns out that due to a bug in the compiler, the code generated for select2 and select3 is not really strict! Moreover, our interval calculation system, which is supposed to detect this kind of error, is quite imperfect and doesn't do it. The solution in these cases is to use min and max to force the arguments to be in the right range of values. For example, to avoid division by 0, you can write 1/max(epsilon, x). Yann Le mar. 24 nov. 2020 à 00:00, yoann.le.bor...@gmail.com < yoann.le.bor...@gmail.com> a écrit : > Hi Florian! > > Le 22/11/2020 à 13:03, Florian Hülsmann a écrit : > > Hi Yoann, > > > > You can simply multiply the functions with their conditions: > > > > f(x) = (x < LowerThreshold) * LowerSignal(x) + ((x >= LowerThreshold) > > & (x <= UpperThreshold)) * LinearSignal(x) + (x > UpperThreshold) * > > UpperSignal(x) > > process = f > > > > Hope this works for you! > > > > Flo > > Unfortunately, this does not work as whatever the signal is, there is > always an attemp to compute log(0) which set the DSP out of order (at > least in FaustIDE). > > > Hi James! > > Le 22/11/2020 à 13:04, James Mckernon a écrit : > > On 11/22/20,yoann.le.bor...@gmail.com <yoann.le.bor...@gmail.com> > wrote: > >> In a non functional language, I would have used a classic if/then/else > >> but, as specified in the faust manual: > >> "WARNING: since select2 is strict (always evaluating both branches), the > >> resulting if does not have the usual "lazy" semantic of the C if form, > >> and thus cannot be used to protect against forbidden computations like > >> division-by-zero for instance." > > I'm actually not sure this is correct; perhaps it's out of date? > > select2 compiles to the ternary operator in C (?:), which I believe is > > lazy in the sense of only evaluating the relevant branch. > > > > The only complication is that faust is 'eager' in evaluating > > expressions at compile-time, where it can, which makes this tricky to > > test. For this reason, select2 seems eager (i.e. it seems to eval both > > branches) when you test it with constant expressions. > > > > process = select2(0, 1/0, 1); // divide by zero error at compile time > > process = select2(1, 1/0, 1); // the same > > process = select2(0, 1/_, 1); // NaN at runtime > > process = select2(1, 1/_, 1); // fine; 1 at runtime > > > > The latter two evaluate a function on faust's input. When testing in > > faust2plot, this is set to an impulse of 1 for the first sample and > > zeroes thereafter. But doing it this way means that faust won't try to > > precompute the value and get the divide by zero at compile-time. > > > > So I believe your waveshaping should just work. I say go ahead and try > it. > > > > Hope that helps, > > James > > I've tested with a select3 and got it working. While, as with Florian > suggestion, log(0) seems to be computed it is handled transparently and > for my personal curiosity I'll also try to use ba.selectoutn and look > under the hood how both are implemented in C. > > Thanks a lot to both of you! > Yoann > > > > > > > _______________________________________________ > Faudiostream-users mailing list > Faudiostream-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/faudiostream-users >
_______________________________________________ Faudiostream-users mailing list Faudiostream-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/faudiostream-users
