[Bug tree-optimization/127314] [middle-end] do-while entry test re-reads a byte that a PRE-hoisted lut lookup already loaded — one more memory access than the source performs (all targets, -O2/-O3)

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127314

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2026-09-11

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
The loop header copying pass before vectorization rotates the loop, exposing
the additional redundancy but there's no further PRE pass which could eliminate
it.  DOM and FRE both cannot handle the case where the redundancy is split to
two
paths via a PHI of the address.

There's duplicate bugs about this, but it's really a fundamental algorithmic
limitation (by design).

RTL PRE could catch this, but it's incredibly weak.

Confirmed for the pass ordering issue.