Re: Octave and GSOC'2008

[John W. Eaton]

On 26-Feb-2008, John Swensen wrote:

| I have a suggestion for a project idea, based on previous mailing list  
| postings.  A while back, Muthiah Annamalai suggested using LLVM to  
| implement JIT for Octave.  John Eaton commented how it would be a big  
| project, mostly due to type inferencing.  I was wondering if it would  
| be a reasonable as a project for GSOC to integrate LLVM in a very  
| limited sense, i.e. only providing JIT for integer indexed loops.   
| While this may seem like a very limited use of LLVM, it would probably  
| provide key insights into how something like LLVM could be integrated  
| more fully into Octave, and it seems like a project that could be  
| completed in a summer's worth of work.

What do you mean by just doing JIT compiling for integer indexed
loops?  How would you do "partial JIT compiling"?  The code that
manages the loop counter itself is already in C++.  The slow part is
interpreting the instructions in the loop body, and storing the
counter in the scripting language variable.  I just don't see how you
can speed that up without also being able to compile the loop body.

In any case, I think maybe we should aim for a little lower level of
difficulty for a summer student project, and one where there could be
some immediate and almost guaranteed success.  For example, there are
still some core Matlab functions in Octave Forge that could be adapted
and moved to Octave itself, and many more core Matlab functions that
are still unavailable in Octave.  I think a student would have a
better shot at success with a project like that, and it is also more
likely that the pool of potential mentors is larger.

jwe