On Thu, Nov 15, 2007 at 09:00:18AM +0000, Simon Peyton-Jones wrote: > Simon, Ian > > We don't have a good way to test for performance regressions, and we really, > really ought to. > > Printing heap allocated, and regarding the test as failed if it get more than > 2% worse might be a simple way. I'd like to have a primitive instruction > count too, but that seems harder. Measuring time is no good at all. > > Quite a lot of bug reports relate to performance, and we just don't have a > good way to add regression tests for them.
Perhaps instrument the bytecode interpreter? This wouldn't help with NCG performance (obviously), but might do a fairly good job at catching Simplifier regressions. Stefan
signature.asc
Description: Digital signature
_______________________________________________ Cvs-ghc mailing list [email protected] http://www.haskell.org/mailman/listinfo/cvs-ghc
