Re: memref_conflict_p

[Michael Hayes <m dot hayes at elec dot canterbury dot ac dot nz>]

Jeffrey A Law writes:
 > Why not just remove the n_refs arg (it's always zero) and only have
 > cases for the PRE_*.?

Yes, it's always zero in this case, but I culled the function from an
optimization pass that I'm about to submit to the egcs project for
dual pack architectures.  This pass requires the n_refs argument.

 > It also seems odd that every case statement sets "offset" or returns, yet
 > you have this code after the switch
 > 
 >   if (offset)
 >     ...
 >   else
 >     ....
 > If you envision having cases where one of the cases does not set
 > offset, then you should also initialize offset instead of leaving 
 > it uninitialized.

Oh, this is just to handle the cases where offset is zero.  Cases
where offset is not set return immediately.

 > Can you fix up those minor issues and resubmit?  (I believe it'll be OK when
 > you resubmit, but I'd like to peek at it again).

I'd like to make the addr_side_effect_eval function global---alias.c
should be a good home.

Michael.