This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: More on c-gimplify.c:gimplify_decl_stmt

On Mon, Jun 07, 2004 at 12:56:24PM -0400, Richard Kenner wrote:
>     > Saying that the first reference uses field F1 of a type and the second
>     > reference uses field F2 of that same type says they can't alias.
>     True, but only for structures (as opposed to unions or variant records).
>     And for structures, the offsets are unique.
> I don't follow.  We're talking about a nest of references and there are
> intermediate fields.

I don't follow.  Sure there's a nest of references.  So what?  I don't
see how that affects my argument at all.

>     Not at all.  We've *got* a biased index -- the offset into the record.
>     We can strength reduce on offset 15+4*i just as easy as we can for [i].
> For strength-reduction, sure.  But I was thinking more of the LNO-style
> optimizations.

The scalar evolution code on the LNO branch can handle a simple linear
equation such as 15+4*i ({15,+,4}) just as easily as a simple direct
index ({0,+,1}).  We are not losing any functionality there.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]