mark_set_resources does not respect UNSPEC_VOLATILE

[Richard Henderson]

On Thu, May 06, 1999 at 08:38:17PM -0700, Mark Mitchell wrote:
> Wow, I agree.  I just copied the code above, but I agree that this is
> not right.  In fact, it looks like unch_memory was similarly abused in
> one case.  How about this patch?

Looks good. 

I had to search a bit before deciding that unch_memory should in fact
accumulate, instead of indicating that the rtx contained _only_ unch_memory.
But the only use is in resource_conflicts_p, and having the bit set when
it needn't be is the conservative choice.

Long term it would be good to scrap most of this crap.  It's not nearly
as accurate as dependancies calculated elsewhere, such as life_analysis
or haifa.  Oh, here's a mind-boggling flash of brilliance, maybe we could
integrate reorg into sched.  Or at least reusing the log-links.


r~