Question on aggregates and GIMPLE

[Richard Kenner]

    Please just emit the __builtin_memcmp yourself in the Ada front end.

That's a bit extreme given that there is plenty of code in the RTL
generators to do these operations and there's no hope of doing any
folding for them if they are buried in builtins.

I thought the whole point of the tree-ssa project was to do things at
a higher level (tree vs. RTL) so that optimizers could function better.
If we go down the direction you suggest, we're making things much *worse*
than it used to be, not better.

I fail to see the problem is having this exposed to the optimizer.
It has nothing to do with vectorization.