[Bug rtl-optimization/85390] possible missed optimisation / regression from 6.3 with conditional expression

[Bug rtl-optimization/85390] possible missed optimisation / regression from 6.3 with conditional expression

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85390

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
More likely we don't normally pull loads above conditionals at all especially
when there is one on each side.
Doing this:
extern int a, b, c;

int f(int x, int b, int a)
{
    __builtin_prefetch((void *) (x ? a : b));
    return c;
}
Shows that.

[Bug tree-optimization/85390] [9/10/11/12 Regression] possible missed optimisation / regression from 6.3 with conditional expression

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85390

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |7.1.0
   Last reconfirmed|                            |2021-12-16
   Target Milestone|---                         |9.5
           Keywords|                            |needs-bisection
     Ever confirmed|0                           |1
            Summary|possible missed             |[9/10/11/12 Regression]
                   |optimisation / regression   |possible missed
                   |from 6.3 with conditional   |optimisation / regression
                   |expression                  |from 6.3 with conditional
                   |                            |expression
                 CC|                            |pinskia at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW
          Component|rtl-optimization            |tree-optimization
      Known to work|                            |6.4.0

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Ok, the issue is just dumb luck:

Take:
extern int a, b, c;

int f(int x)
{
    __builtin_prefetch((void *) (__SIZE_TYPE__)(x ? a : b));
    return c;
}

int f1(int x)
{
    int t = x ? a :b;
    __builtin_prefetch((void *)(__SIZE_TYPE__)t);
    return c;
}

The f1 uses cmov while f does not.

The gimple level difference is there is two cast (one in each branch) for f
while only one cast outside of the PHI node.
But that was that way even in GCC 6.
I wonder what changed in GCC 7 which caused this issue to appear.

[Bug tree-optimization/85390] [9/10/11/12 Regression] possible missed optimisation / regression from 6.3 with conditional expression

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85390

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug tree-optimization/85390] [9/10/11/12 Regression] possible missed optimisation / regression from 6.3 with conditional expression

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85390

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |marxin at gcc dot gnu.org
           Keywords|needs-bisection             |

--- Comment #4 from Martin Liška <marxin at gcc dot gnu.org> ---
> But that was that way even in GCC 6.
> I wonder what changed in GCC 7 which caused this issue to appear.

Changed with r7-2128-gebe8f3226c941730

[Bug tree-optimization/85390] [10/11/12/13 Regression] possible missed optimisation / regression from 6.3 with conditional expression

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85390

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|9.5                         |10.4

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 9 branch is being closed

[Bug tree-optimization/85390] [10/11/12/13 Regression] possible missed optimisation / regression from 6.3 with conditional expression

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85390

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|10.4                        |10.5

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 10.4 is being released, retargeting bugs to GCC 10.5.

[Bug tree-optimization/85390] [10/11/12/13/14 Regression] possible missed optimisation / regression from 6.3 with conditional expression

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85390

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|                            |14.0
   Target Milestone|10.5                        |14.0
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed fully for GCC 14 by r14-573-g69f1a8af45d8a4 which pulls out the cast from
the branches and allows ifcvt to happen.