On Mon, Sep 18, 2006 at 12:09:36PM -0700, Mark Mitchell wrote: > Mike Stump wrote: > >On Sep 15, 2006, at 2:32 PM, Ross Ridge wrote: > >>Also, I don't think it's safe if you merge only functions in COMDAT > >>sections. > > > >Sure it is, one just needs to merge them as: > > > >variant1: nop > >variant2: nop > >variant3: nop > > [ ... ] > > > >this way important inequalities still work. > > Yes, that will work. But, in practice, people will also want the mode > where you do not insert the nops, and just accept that some functions > compare equal when they shouldn't. So, we should have a switch for that > mode too.
It certainly is true that if we generate one set of routines for vector<int>, vector<long>, and vector<T*> for all T, a program that uses them all could have a lot of nops. But there's another possibility: we can implement the variants as inline functions that call the common implementation. If the address of the function is not taken in a given file, we can make a direct call. If the address is taken, we need an "outlined" copy of the function. These out-of-line copies can be handled in the usual linkonce/COMDAT way, so that they resolve to one copy. So it seems that just having address-taking block inlining suffices.
