ipa-cp drawback

[Ian Lance Taylor]

charfi asma <charfiasma@yahoo.fr> writes:

> I have question concernais the constant propagation drawbacks.
> I
> red in one of the GCC summit paper that Constant propagation is limited
> to create at most one clone of each function body and only in the case:
> - the operand is the same constant in all calls to the function 
> - the operand would not be propagated without cloning because function might be called externally
> This makes the interaction with the inliner suboptimal
>
> Could  so explain more this idea? did this  means that ipa-cp will produce only one version of a calle function? 
> Could so  please give me an example that shows this drawbak? 

I think that the implementation of IPA CP has been improved since that
paper was written.

Ian