On 16/07/2008, at 9:01 PM, Neil Mitchell wrote:
I have also previously talked with Ben Lippmeier about using SYB in DCC (http://www.haskell.org/haskellwiki/DDC) and was told the performance loss was "too great" and he had to switch back. However, these are merely anecdotes.
It was about 3 times slower - though DDC does lots of little passes over the various ASTs, so its fairly susceptible to problems with traversal speed.
BTW: Hi John, Simons. Feel free to throw any problems wrt register allocation my way that might be exposed by new work on the back end. Is the plan still to use the graph libraries to help with allocating stack slots?
If the back-end work manages to increase register pressure, or otherwise starts producing bigger graphs, then I'll have another look at the performance of the allocator code to make sure we don't loose anything against the old linear allocator.
Ben. _______________________________________________ Cvs-ghc mailing list [email protected] http://www.haskell.org/mailman/listinfo/cvs-ghc
