[Bug tree-optimization/102511] [12 Regression] GCC produces incorrect code for -O3: first element of the array is skipped after r12-3903

["aldyh at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102511

--- Comment #10 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #7)
> (In reply to Aldy Hernandez from comment #6)
> > Describing the process to get here makes it abundantly clear that we need to
> > improve the process of debugging this.  We need a way to turn on the solver
> > debugging from the command line (--param??), and not by some magic #define. 
> > Also, some counter that matches the path being registered with the
> > equivalent solver dump would be useful.  This way someone could easily find
> > the problematic thread in the solver dump.  I'll look into this.
> 
> Maybe look into dbgcnt.def and their usage. It should help with the counter
> situtation :).

Thanks Andrew.  I've used your idea in my committed patch here:

https://gcc.gnu.org/pipermail/gcc-patches/2021-September/580369.html

It's not perfect, but it should help in diagnosing these issues in the future. 
I hope to improve the dumps in the next month, to help any stage3 bugfixing.