Re: [trans-mem] cgraph edges vs function cloning

[Martin Jambor <mjambor at suse dot cz>]

Hi,

On Tue, Jul 28, 2009 at 04:26:12PM -0700, Richard Henderson wrote:
> On 07/28/2009 10:44 AM, Richard Henderson wrote:
> >I guess I'll poke at cleaning this up today. I've got to
> >familiarize myself with how virtual clones work...
> 
> The virtual clones that ipa-cp makes seems to be easy.
> 
> My thought here is that since (virtual) clones don't
> have actual bodies (and when they acquire bodies they
> cease to be clones), then there's no reason for them
> to have callee edges at all. 

That is not really true.  Consider the following example:

static void a (int i)
{
  DO_SOMETHING_WITH_I (i);
}

void b (int i)  /* Not static! */
{
  a (i);
}

void c (void)
{
  b (2);
}

After ipa-cp (even today), we might end up with a call graph where c
would call b.clone.0 which in turn would call a.clone.1, while the
original b would still call the original a.

Moreover, I have an ipa-cp improvemant patch that removes callee edges
of virtual clones if it can prove that they are never called after a
constant is substituted into a parameter, like in the example below:

int f(int i)
{
  if (i == 0)
    {
      call_some_nasty_function(...);
    }
  ...
}

int g() {return f(1);}

Martin