[EMAIL PROTECTED] wrote on 07/08/2006 17:35:17: > Hi Ben, > > It's nice to know what opcodes are popular! However, sorting them in > whatever order within a switch will likely not make a difference. > This switch will be converted to a jumptable, so the compiler will > order the cases again. > > cheers > Eric >
Hi Eric, It probably won't, but there appeared to be some desire for this on irc, and this is the kind of project I can fit into my lunch break! I think it's worth a try to see if it makes a difference. Unfortunately, I can't test it here. Cheers, Ben > > On Aug 7, 2006, at 4:35 PM, [EMAIL PROTECTED] wrote: > > > Hi PyPys! > > > > Here's a patch that counts the number of times the different > > bytecodes are > > used. I haven't tried doing translation yet as I can't use the work > > machine cycles, but it can't be far off. I've put the counts on the > > object > > space as this is accessable from most places. > > > > I've also implemented something that should resort the opcodes in the > > switch statement of the bytecode dispatcher according to the > > statistics I > > found. This may have a slight speed improvement. > > > > > > > > Cheers, > > Ben > > <diff.txt> > > _______________________________________________ > > [email protected] > > http://codespeak.net/mailman/listinfo/pypy-dev > > _______________________________________________ > [email protected] > http://codespeak.net/mailman/listinfo/pypy-dev > _______________________________________________ [email protected] http://codespeak.net/mailman/listinfo/pypy-dev
