Re: Natural loop detection

[Richard Henderson <rth at cygnus dot com>]

On Mon, Nov 22, 1999 at 10:26:13AM +1300, Michael Hayes wrote:
> I originally had it in loop.c but thought it should live with the CFG
> stuff so that it could be used generically and not get hijacked by the
> loop optimizer.

I don't see that having the code in loop.c implies that it
will be hijacked.  My rationale for putting the code in loop.c
is that it's loop related.  *shrug*

>  > I prefer using basic block pointers over indicies, even
>  > if it does take up slightly more room on 64-bit hosts.
> 
> What is the rationale here?  Jan Hubica warned me yesterday that you
> wouldn't like this ;-) 

Mostly that it feels more like a "real" data structure.  It
also avoids unnecessary indirection.

> I was thinking of replacing the exits bitmap with lists of exiting
> edges anyway since this is probably easier to use.

That would be ok too.


r~