Re: PR 2828: Following up

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Richard" == Richard Henderson <rth@redhat.com> writes:

    Richard> I suspect that (without cond_exec and its odd starting
    Richard> conditions) virtually all of the time things settle down
    Richard> to their proper form rather rapidly.  So while it may not
    Richard> be guaranteed to terminate, we've never noticed before.

:-)

    Richard> Does your enable checking bit pass if, while queuing the
    Richard> initial set of blocks to check, you clear each block's
    Richard> global_live_at_start?  Oh, no, it needn't.  
    
Been there, though that.

    Richard> Well, perhaps the check could notice that this is the
    Richard> first time the block has been processed this invocation
    Richard> of global_live_at_start.  I would expect all of the
    Richard> register deaths due to code changes to happen then.

Yes, I was trying to convince myself of that.  Haven't done it yet,
but it might work.  My brain hurts, and there are lots of dots of
arrows on my whiteboard.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com