Re: Announce: scnim - Write SuperCollider UGens in Nim (Proof of Concept)

>Hi Carlo, 
>
>this is an exciting project, thanks for sharing. Nim seems to be an 
>interesting language as well. A while ago, I thought about something 
>like this with Scala Native (http://www.scala-native.org/en/latest/), 
>although I had considered rather to see how to difficult it would be to 
>implement part of the server API, rather than writing UGens that could 
>then be linked back into scsynth. Your documentation will be very useful 
>for similar attempts. Have you gathered any insight into the performance 
>of using Nim for the UGens?
>
>best, .h.h. 

Pleasure!

Nim is a bit different than most contenders because it actually generates
(at your option) C or C++ code, and that code is fairly readable. Below is
Nim's C++ result of the "next" procedure. The scary-looking stuff at the
bottom is just explicit type casts that Nim insist you do, and those have
zero impact. The rest is just a loop in C, and those have been tinkered with
by compiler optimizers for 30 years. So I think we can safely say that if
you use the same rules to write UGens in Nim compared to C++ (don't allocate
memory, use linear data structures like arrays), there is no (zero, none)
performance hit.

float* input = (*(*unit).mInBuf);
float* output = (*(*unit).mOutBuf);
  NI32 i;
  i = (NI32)0;
  NI32 res = ((NI32) 0);
  {
    while (1) {
      if (!(res <= numSamples)) goto LA3;
      i = res;
      (*((float*) ((NI)((NU64)(((NI) (ptrdiff_t) (output))) +
(NU64)((NI)(((NI) (i)) * ((NI) 4))))))) = (*float*) ((NI)((NU64)(((NI)
(ptrdiff_t) (input))) + (NU64)((NI)(((NI) (i)) * ((NI) 4)))))));
      res += ((NI) 1);
    } LA3: ;
  }
}

The same applies to sharing data- in this case, the input buffer is created
in scsynth in C++, passed to Nim-generated C++ code, worked on, and
returned- nothing is copied anywhere or restructured or anything. You can
look at Nim as tool to help you write efficient C or C++ code. The price you
pay is in build time and understanding how to write the interface- but once
it's all compiled, you get the same CPU instructions and the same memory
layout, so you get the same performance.





--
Sent from: http://new-supercollider-mailing-lists-forums-use-these.2681727.n2.nabble.com/SuperCollider-Developers-New-Use-this-f2681767.html

_______________________________________________
sc-dev mailing list

info (subscription, etc.): http://www.birmingham.ac.uk/facilities/ea-studios/research/supercollider/mailinglist.aspx
archive: http://www.listarc.bham.ac.uk/marchives/sc-dev/
search: http://www.listarc.bham.ac.uk/lists/sc-dev/search/