[Bug rtl-optimization/57676] [4.9/5/6 Regression] ICE: Maximum number of LRA constraint passes is achieved (30)

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57676

--- Comment #10 from Vladimir Makarov <vmakarov at gcc dot gnu.org> ---
(In reply to Jeffrey A. Law from comment #9)
> Vlad, would it make sense to record what insns needed a reload & what insns
> were generated for reloads.  Then on the next iteration, if the insns
> needing a reload are different than the ones needing reloads in previous
> iterations + insns generated for reloads in previous iterations, allow
> iteration to continue.
> 
> ie, detect the case when we've got this kind of cascading effect and we're
> still making progress and as long as we're still making progress allow
> iteration to continue.

Yes, it makes a perfect sense.  I should implement this.  I'll think how to do
it better.  But I think realistically this work can be done only for GCC7.

Thanks for providing a fresh view how to handle this problem.