https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104950
Drea Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |pinskia at gcc dot
gnu.org
Status|NEW |ASSIGNED
Depends on| |125896
Component|rtl-optimization |tree-optimization
--- Comment #7 from Drea Pinski <pinskia at gcc dot gnu.org> ---
-O2 we have:
```
if (_51 < _52)
goto <bb 7>; [50.00%]
else
goto <bb 6>; [50.00%]
<bb 6> [local count: 536870912]:
# VUSE <.MEM_8(D)>
_49 = MEM[(const struct Node *)_57].l;
goto <bb 8>; [100.00%]
<bb 7> [local count: 536870912]:
# VUSE <.MEM_8(D)>
_50 = MEM[(const struct Node *)_57].r;
<bb 8> [local count: 1073741824]:
# _48 = PHI <_50(7), _49(6)>
```
-O3 is messed up do to split paths too.
Note in the reduced testcase in comment #1 we can hoist the loads too:
Hoisting adjacent loads from 4 and 3 into 2:
# VUSE <.MEM_3(D)>
_5 = a_4(D)->l;
# VUSE <.MEM_3(D)>
_6 = a_4(D)->r;
So I need to look into why we can do it for that but not for the original case.
(In reply to Hongtao.liu from comment #4)
> Not for this one
>
> float
> foo (float a, float b, float *c, int i, int j)
> {
> return a > b ? c[i] : c[j];
> }
That is handled on the trunk already.
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125896
[Bug 125896] load common does not support non-only MEM_REF-refences
