[Bug tree-optimization/89491] Inline jump tables

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89491

Andrew Pinski  changed:

   What|Removed |Added

   Last reconfirmed||2021-07-19
 Status|UNCONFIRMED |NEW
 Ever confirmed|0   |1
   Severity|normal  |enhancement
   Keywords||missed-optimization

--- Comment #6 from Andrew Pinski  ---
The big question is how useful is this in real code?
ICC does it, then there must have been a benchmark where this improves but I
don't know which one.

Also it looks like ICC does:
int f(int x)
{
  const p *t = arr[x];
  if (t == square)
return square(1);
  else
return add(1);
}

[Bug tree-optimization/89491] Inline jump tables

[david.bolvansky at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89491

--- Comment #5 from Dávid Bolvanský  ---
Let's take the original example with small modification:

int square(int x) { return x*x; }
int add(int x) { return x + x; }
typedef int (*p)(int);
static const p arr[4] = {square, add};
int test(int x) {
int res = arr[x](1);
return res;
}

Nothing is expanded/inlined. ICC can do it for "two items in arr" case.

[Bug tree-optimization/89491] Inline jump tables

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89491

Jakub Jelinek  changed:

   What|Removed |Added

 CC||jakub at gcc dot gnu.org

--- Comment #4 from Jakub Jelinek  ---
"Inlining" from vtables is called devirtualization, and GCC performs that
either if there is just a single possibility, or with PGO if a particular
method is known to be very common.  Doing it in other cases is probably not a
very good idea.

[Bug tree-optimization/89491] Inline jump tables

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89491

--- Comment #3 from Andrew Pinski  ---
(In reply to Dávid Bolvanský from comment #2)
> Right, static helps.
> 
> What about more complex examples, like inlining vtables?
> https://gcc.godbolt.org/z/ZXkRYa

Your example is much more complex.  The vtable might be constant but where the
vtable is stored is not.

[Bug tree-optimization/89491] Inline jump tables

[david.bolvansky at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89491

--- Comment #2 from Dávid Bolvanský  ---
Right, static helps.

What about more complex examples, like inlining vtables?
https://gcc.godbolt.org/z/ZXkRYa

[Bug tree-optimization/89491] Inline jump tables

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89491

--- Comment #1 from Andrew Pinski  ---
arr is non-constant so GCC cannot assume it would be square or add all the
time.