[Bug rtl-optimization/82524] [7/8 Regression] expensive-optimizations produces wrong results

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82524

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 42346
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=42346&action=edit
gcc8-pr82524.patch

Perhaps it could, but I think such a change isn't at least backportable and not
sure how much work would that be, after all, I think LRA usually works with
insn operands, not RTL that wraps them, so this would need to be an exception.

Here is an attempt to adjust the affected patterns, some of them with
match_dups, others with rtx_equal_p in conditions.

Let me try to write some testcase that attempts to test all the patterns
whether they can be matched if there is matching target.