I know we have some old(er) Apple developers out there, so I wanted to ask:
does MakeDataExecutable have a particular penalty?
I'm working on a PowerPC nanojit for Firefox so that Firefox's JavaScript
tracer can emit PPC instructions. It works, but it takes a huge initial
penalty, more than I would have expected. The pure-C++ JavaScript interpreter
beats it on many tasks. I'm sure that part of that is trying to beat gcc's
very good PPC optimizer, but I want to eliminate other variables.
It also seemed that PPC mulli is still slow compared to unrolled naked adds
and bitshifts, and naturally the Firefox bytecodes do not have support for
fused multiply-add or other things that POWER does well.
--
------------------------------------ personal:
http://www.cameronkaiser.com/ --
Cameron Kaiser * Floodgap Systems *
www.floodgap.com * ckaiser at
floodgap.com
-- "In God We Trust (All Others We Monitor)" ----------------------------------