https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111560
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |rguenth at gcc dot gnu.org
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
Last reconfirmed| |2023-09-24
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed. Simpler testcase:
int e,f;
void test(int a, int b, int c, int d)
{
e=a+b+c; //line 5
f=d+b+c; //"b+c" can be replaced with the value at line 5
}
it's as Andrew says plus re-associating for global optimal CSE is difficult,
imagine adding
g = a + b + d;
to the above. There's either a + b in common with e but then that breaks
commoning b + c in e and f or there is b + d in common with f also breaking
the other CSE.
Our re-association only produces a canonical order within a single expression.
That doesn't by design allow CSE of common subexpressions in other expressions.
I don't know of any work (paper) describing how to attack this global
optimization problem.