[Bug tree-optimization/107823] [13 Regression] Dead Code Elimination Regression at -Os (trunk vs. 12.2.0)

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107823

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|rguenth at gcc dot gnu.org         |unassigned at gcc dot gnu.org
             Status|ASSIGNED                    |NEW
                 CC|                            |aldyh at gcc dot gnu.org

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think it's working as intended now - in GCC 12 we did

Checking profitability of path (backwards):  bb:4 (4 insns) bb:3 (latch)
  Control statement insns: 2
  Overall: 2 insns
  [4] Registering jump thread: (3, 4) incoming edge;  (4, 5) nocopy;
path: 3->4->5 SUCCESS

while on trunk we have

Checking profitability of path (backwards):  bb:4 (4 insns) bb:3 (latch)
  Control statement insns: 2
  Overall: 2 insns

Checking profitability of path (backwards):   FAIL: Jump-thread path not
considered: duplication of 2 insns is needed and optimizing for size.
path: 3->4->xx REJECTED

so the fix works exactly as intended.

The pattern of a loop where the control stmt is statically resolvable
both in the first and the second iteration would likely need special
heuristical treatment, allowing to see the full 2->4->3->4->5 thread
and anticipating elimination of the conditionals and the original
block copies.  But the machinery isn't set up for this at all.

It works just fine at -O2.