Le 26 mai 2013 à 00:14, Marc Joliet <mar...@gmx.de> a écrit :
> Am Sat, 25 May 2013 08:59:58 +0200
> schrieb Stéphane Letz <l...@grame.fr>:
>
> [...]
>>> Some obvious alternatives to the CFFI would be:
>>>
>>> - Cython, which is, to my understanding, not dynamic enough for what I
>>> wanted,
>>> but which can be used to wrap C++ code, so would have the advantage of not
>>> requiring FAUST2.
>>>
>>> - the FAUST2 LLVM backend + Unladen Swallow, which ties you to CPython,
>>> whereas
>>> the CFFI works (or at least will work) with various Python implementations.
>>>
>>> - Ctypes, which is dynamic, like the CFFI, but supposedly not as nice to use
>>> and slower.
> [...]
>>
>> Have you ever thought about using the *dynamic* Faust technology in this
>> application? That is libfaust + LLVM?
>>
>> Stéphane
>
> Hmmm, no, I hadn't, because 1.) I hadn't really heard of libfaust before
> outside
> of the Makefile and stupidly assumed it was an internal library (perhaps a
> more
> prominent mention should be made somewhere in the docs?), and 2.) I sort of
> took my first idea for an implementation and ran with it, just to get
> something
> working quickly.
>
> Now that you mentioned it, I have looked into it more and it looks like it
> won't
> change much, other than that I don't have to spawn a faust process anymore,
> i.e.
>
> dsp code -> faust2 -> C code -> CFFI -> Python object
>
> becomes
>
> dsp code -> libfaust -> C code -> CFFI -> Python object
The idea would possibly be:
dsp code -> libfaust -> LLVM IR -> CFFI (??) -> Python object
>
> However, you also mentioned LLVM: do I understand correctly that you mean
> using
> the LLVM backend as (I think) it is used in Pure? That is, generate LLVM
> bitcode and link that to Unladen Swallow/CPython (however that exactly
> happens)? I dismissed it since that ties you to CPython 3.3+, as I mentioned
> in my original email. Plus I like having a pure python solution :) .
>
> Anyway, after lots of fiddling with compiler/Makefile.unix [0] and learning
> about the "-rpath" linker option I eventually got libfaust.so and my test
> program compile without messing with my system's LDPATH and have now tried it
> out in a local branch following the scheme above.
>
> While I think that I prefer using libfaust to spawning a faust process (it
> feels more elegant, and the FAUST class and the instantiation overhead shrunk
> a
> bit), I noticed one important (strange) difference between the two: the order
> of meta-data declarations (i.e., the line order in the header and the order of
> ->declare() calls) changes between calls to compile_faust(). This renders the
> cffi's caching mechanism useless and leads to re-compiling on every
> instantiation. Simple inline code strings still cache fine, as long as they
> don't contain more than one meta-data declaration. Here's an example:
>
> $ diff -U4 test{1,2}.txt
> --- test1.txt 2013-05-25 20:02:38.152409591 +0200
> +++ test2.txt 2013-05-25 20:02:34.688500257 +0200
> @@ -34,12 +34,12 @@
> void* uiInterface;
> } UIGlue;
>
> //-----------------------------------------------------
> -// version: "0.1"
> // name: "Dattoro notch filter and resonator (Regalia)"
> // license: "MIT"
> // copyright: "(c)Marc Joliet 2013"
> +// version: "0.1"
> // author: "Marc Joliet"
> //
> // Code generated with Faust 2.0.a9 (http://faust.grame.fr)
> //-----------------------------------------------------
> @@ -84,14 +84,13 @@
> free(dsp);
> }
>
> void metadatamydsp(MetaGlue* m) {
> - m->declare(m->mInterface, "version", "0.1");
> m->declare(m->mInterface, "name", "Dattoro notch filter and resonator
> (Regalia)");
> - m->declare(m->mInterface, "filter.lib/name", "Faust Filter
> Library");
> - m->declare(m->mInterface, "filter.lib/license", "STK-4.3");
> - m->declare(m->mInterface, "filter.lib/reference",
> "https://ccrma.stanford.edu/~jos/filters/";); m->declare(m->mInterface,
> "license", "MIT");
> + m->declare(m->mInterface, "copyright", "(c)Marc Joliet 2013");
> + m->declare(m->mInterface, "version", "0.1");
> + m->declare(m->mInterface, "filter.lib/reference",
> "https://ccrma.stanford.edu/~jos/filters/";); m->declare(m->mInterface,
> "math.lib/name", "Math Library"); m->declare(m->mInterface,
> "math.lib/author", "GRAME"); m->declare(m->mInterface,
> "math.lib/copyright", "GRAME"); m->declare(m->mInterface,
> "math.lib/version", "1.0"); @@ -100,13 +99,14 @@
> m->declare(m->mInterface, "music.lib/author", "GRAME");
> m->declare(m->mInterface, "music.lib/copyright", "GRAME");
> m->declare(m->mInterface, "music.lib/version", "1.0");
> m->declare(m->mInterface, "music.lib/license", "LGPL with exception");
> - m->declare(m->mInterface, "copyright", "(c)Marc Joliet 2013");
> - m->declare(m->mInterface, "filter.lib/author", "Julius O. Smith
> (jos at ccrma.stanford.edu)");
> - m->declare(m->mInterface, "filter.lib/copyright", "Julius O. Smith
> III"); m->declare(m->mInterface, "filter.lib/version", "1.29");
> + m->declare(m->mInterface, "filter.lib/license", "STK-4.3");
> m->declare(m->mInterface, "author", "Marc Joliet");
> + m->declare(m->mInterface, "filter.lib/name", "Faust Filter
> Library");
> + m->declare(m->mInterface, "filter.lib/author", "Julius O. Smith
> (jos at ccrma.stanford.edu)");
> + m->declare(m->mInterface, "filter.lib/copyright", "Julius O. Smith
> III"); }
>
> int getSampleRatemydsp(mydsp* dsp) { return dsp->fSamplingFreq; }
>
> I don't understand why that happens when using libfaust but not faust, since
> faust merely calls compile_faust_internal() :-/ .
Hum.. interesting! I'll have a look.
>
> [0] Gentoo's libtool is quite a bit newer than the one on Mac OS X and has
> changed quite a bit, the libtool-1.3.5 package (the version I think OS X has)
> only installs libltdl, and it appears the libtool developers *really*,
> *really*
> want you to use automake, so I gave up and basically had the Makefile call
> "g++
> -shared" instead.
Stéphane
------------------------------------------------------------------------------
Try New Relic Now & We'll Send You this Cool Shirt
New Relic is the only SaaS-based application performance monitoring service
that delivers powerful full stack analytics. Optimize and monitor your
browser, app, & servers with just a few lines of code. Try New Relic
and get this awesome Nerd Life shirt! http://p.sf.net/sfu/newrelic_d2d_may
_______________________________________________
Faudiostream-users mailing list
Faudiostream-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/faudiostream-users
