[Bug tree-optimization/68671] [5/6 Regression] gcc.dg/torture/pr66952.c FAILs with -fno-tree-dce

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68671

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
no_side_effect_bb should already ensure that the bbs that
maybe_optimize_range_tests is considering don't contain anything but the final
condition and optionally non-trapping no-side-effects assignments which are
only used within the same bb (so pretty much to compute the condition, or have
no non-debug uses at all).
So, I believe we just need to go and reset value ranges for the SSA_NAMEs in
the blocks after the first one we've changed up to the last bb considered,
because changes to the conditions or SSA_NAMEs that feed those conditions could
cause non-trapping no-side-effect assignments in those bbs to be executable
even when they wouldn't be in the original program.