--with-computed-gotos optimizations probably nullified by cse

[Michael Hudson-Doyle]

The packaging for python 3.3 enables the --with-computed-gotos flag to configure. But as discussed in the gcc documentation (http://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-fgcse-781) the benefits of computed gotos are often eliminated by common subexpression elimination.

Testing I have done, which I need to be more systematic about, certainly suggests that this is the case: if you pass --with-computed-gotos and arrange to include -fno-gcse in the optimization flags, you see performance improvements. If you don't arrange to include -fno-gcse, you get performance identical to not passing --with-computed-gotos.

It would be best to just disable CSE for PyEval_EvalFrameEx, but that function is so dominant in run time that disabling it globally is and getting the benefits of the computed gotos is probably still a win overall.