Re: [IRA] Fix PR rtl-optimization/45912

[Eric Botcazou <ebotcazou at adacore dot com>]

> So I'd like to consider your previous patch as a temporary solution
> (although some temporary solutions stay forever :).  I think that
> solution I wrote here is better but not ideal too (for example in a
> situation when a cover class is union of some other classes).  I'll put
> this work on my TODO list but honestly don't know when I start working
> on this.

OK, thanks for the explanation.

Btw, there is a strange loop in setup_class_hard_regs:

      for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
	{
	  ira_non_ordered_class_hard_regs[cl][0] = -1;
	  ira_class_hard_reg_index[cl][0] = -1;
	}

-- 
Eric Botcazou