Hi Dario, I am not aware of any recent changes in delay-line interpolation, but there was a change in tau2pole to allow tau to be zero without crashing numerically. To check if this is it, change the definition of tau2pole in basics.lib back to
tau2pole(tau) = exp(-1.0/(tau*ma.SR)); - Julius On Fri, Mar 6, 2020 at 5:10 AM Dario Sanfilippo <sanfilippo.da...@gmail.com> wrote: > > At the moment, this is the quickest test I can do, but there is still a lot > in the network so it's not enough to isolate the problem. > > Julius: has the algorithm for the smooth delay lines changed recently? Those > are also used in my other network that shows the same problem. Otherwise, I > wouldn't know what may have changed. You can see the SVG here: > https://www.dropbox.com/sh/3hx6k4mjabmrult/AACH_Tea05lUxuPVlskevblVa?dl=0. > > Compiling the code with 2.15 and 2.23, I already see the outputs diverging > after one second: > > 2.23: 0.00046198073, -0.000696233998, > 2.15: 0.000461812917, -0.000695491733, > > I think that Stéphane once mentioned that there is a consistency benchmark > that checks if the output of a sequence is close enough to that generated by > previous Faust version. Is that correct? If so, unless some algorithm such as > the interpolation in FDLs changes, wouldn't it be reasonable to expect an > identical output, hence have the benchmark fail if the sequence is not > identical? > > Best, > Dario > > > On Fri, 6 Mar 2020 at 08:29, Stéphane Letz <l...@grame.fr> wrote: >> >> Hi Dario, >> >> We fixed a bug in vector mode code generation (so when using -vec at >> compilation time) in version 2.22.3 february the 4th. Are you using the -vec >> mode when compiling? >> >> If not (that is if you are using default mode which is so called « scalar » >> code), the code should always behave the same. We should really dig into >> this issue. Are you able to prepare the *simplest DSP example you can* that >> causes the issue and send it so me ? >> >> Thanks. >> >> Stéphane >> >> > Le 6 mars 2020 à 00:34, Dario Sanfilippo <sanfilippo.da...@gmail.com> a >> > écrit : >> > >> > Hello, list. >> > >> > I'm noticing that my networks sound different when compiled with different >> > versions of Faust. >> > >> > I'm doing the tests with self-oscillating closed systems that I initialise >> > with a Dirac impulse; there are no randomness or external signals in the >> > network so they are deterministic systems. >> > >> > I think that I have experienced this from 2.18 to 2.20, and now from 2.20 >> > to the current master-dev 2.23.2. >> > >> > Is this is something that you consider possible and that might happen >> > often over new Faust releases? >> > >> > What could change in Faust so that calculations result in different values? >> > >> > Best, >> > Dario >> > _______________________________________________ >> > Faudiostream-users mailing list >> > Faudiostream-users@lists.sourceforge.net >> > https://lists.sourceforge.net/lists/listinfo/faudiostream-users >> -- "Anybody who knows all about nothing knows everything" -- Leonard Susskind _______________________________________________ Faudiostream-users mailing list Faudiostream-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/faudiostream-users
