[pypy-dev] Pypy jit and (meta) genetic algorithms

[Ronny Pfannschmidt]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 09/27/2011 10:43 PM, David Naylor wrote:
> Hi All
> 
> It occurred to me that with the many options available for jit (such as 
> inlining, function_threshold) there may be some merit to optimising those 
> values.  I would expect that the optimised values would be workload specific 
> however if a workload takes days to run then it would be worth optimising.  
> 
> I recall an article that used genetic algorithms to select the best parameters 
> (for gcc) that produces the fastest execution.  Is there an equivalent program 
> for pypy?  Or if it is easy enough could someone put together such a (shell 
> script) program?  
> 
> I, unfortunitely, have no experience with genetic algorithms nor know how to 
> optimise the jit parameters. 
pretty simple genomes could already do i suppose (pyevolve should have
everything you need

the main tricky parts will be choosing what sizes of population and how
to test

i am very sure that this will be very computation-intensive

- -- ronny
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
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=BWpS
-----END PGP SIGNATURE-----