Re: PATCH: PR middle-end/35897: DSE doesn't support targets with wide ?registers

[Kenneth Zadeck <zadeck at naturalbridge dot com>]

H.J. Lu wrote:
> On Fri, Apr 11, 2008 at 10:27:41AM -0400, Kenneth Zadeck wrote:
>   
> > > Hi Kenny,
> > >
> > > My patch is at
> > >
> > > http://gcc.gnu.org/ml/gcc-patches/2008-04/msg00937.html
> > >
> > > Does it look OK?
> > >
> > >   
> > >       
> > sorry, i did not see the entire patch.   it does look ok.
> >     
>
> Hi Kenny,
>
> I have a question on DSE. There are
>
>   store_info->positions_needed = lowpart_bitmask (width);
>   store_info->group_id = group_id;
>   store_info->begin = offset;
>   store_info->end = offset + width;
>
> So
>
>  store_info->begin <= offset < store_info->end
>
> But there are 2 instances of
>
>   && (offset >= store_info->begin)
>   && (offset + width <= store_info->end)
>     {
> ...
>     }
>
> Shouldn't they be
>
>   && (offset >= store_info->begin)
>   && (offset + width < store_info->end)
>     {
> ...
>     }
>
> If I am correct, I can provide a patch.
>
> Thanks
>
>
> H.J.
>   
I do not think that you are correct.  This does not mean that there is 
not some other inconsistancy but i believe that this code is correct.

on line (before your patch) 1375 we have:
 store_info->end = offset + width;

in the two cases that you mention, if offset == store_info->begin, then 
offset + width will be equal to store_info->end for a full word 
operation.   So i think that the code is correct.  

i.e. if offset is 10, begin 10, and width is 1, then end will be 11 but 
offset + width will also be 11.

this is consistent with the loops that are     for (i = 
store_info->begin; i < store_info->end; i++)
will only iterate 1 time - for the value 10.

Kenny