|Seprerating algorithms from implementations (long) TSharp@Serif.com (Toby Sharp) (2000-08-27)|
|Re: Seprerating algorithms from implementations (long) firstname.lastname@example.org (felix) (2000-09-02)|
|Re: Seprerating algorithms from implementations (long) email@example.com (Rob Hutchinson) (2000-09-08)|
|Re: Seprerating algorithms from implementations firstname.lastname@example.org (Tom Moog) (2000-09-09)|
|From:||"Rob Hutchinson" <email@example.com>|
|Date:||8 Sep 2000 01:58:53 -0400|
|Organization:||Posted via Supernews, http://www.supernews.com|
1.)This algorithm description language idea is interesting. Seems
like a huge task though. Seems to me you would need to create both an
algorithm description language AND compilers for this language for
every possible target processor. Even if this was done I still find
it hard to believe that the compilers could optimize to the same
degree that hand coding assembly language gives. It would be a very
big project to create a satisfactory algorithm description language;
i.e. easy to program in yet still capable of specifying all relevant
info the compiler would need. Likewise the creation of even one such
compiler would be a huge task.
2.)My own fantasy in this area would be to create a framework of some
kind, (e.g. a set C++ DSP classes), designed to perform all the basic
signal processing operations. These classes could then be used to
generate all manner of DSP applications. The framework would be
designed such that all time critical portions, (real multiply, complex
multiply, magnitude of a complex number, ... etc.) would be isolated
in a nice simple modular way. Then I could prototype the DSP
application using the C++ classes and when I needed to optimize for a
particular platform I could pull out the time critical module, code it
optimally for the platform, and place it back into the main code body.
The main idea being to minimize the number of lines of code I have to
re-write to optimize for a particular platform. Haven't made much
progress with this yet....
Toby Sharp <TSharp@Serif.com> wrote in message
> We might be able to solve this problem if there were some way to
> describe algorithms apart from their implementations.
Return to the
Search the comp.compilers archives again.