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]


> On Mon, 2004-01-12 at 20:55, Jan Hubicka wrote:
> > Does this seem to make sense?
> > Bootstrapped/regtested i686-pc-gnu
> > Hope I didn't messed up something obvious
> > 
> Thanks Jan.  The patch is fine and fixes PR11761.  I made a couple of
> minor adjustments and checked it in.  It provides some slight code size
> improvements and identical compile times.
> I would like us to address the wrinkle wrt TREE_ADDRESSABLE and arrays
> referenced with non-constant indices.  Perhaps we should have a
> different bitflag to avoid confusing the RTL expanders.  That would
> allow us to remove the call to discover_nonconstant_array_refs.

You can see the original patch implementing the flag.  Once we will find
a need to for sich a fine difference, I think I can just rescuesce it.

> + 
> +   /* Mark arrays indexed with non-constant indices with TREE_ADDRESSABLE.
> +      FIXME: Is this really needed long-term?  This is done for the benefit
> +      of the RTL expanders.  */
> +   discover_nonconstant_array_refs ();

Yes, I don't see any reason teaching RTL expanders to maintain real
arrays (used with nonconstant indices) in registers.


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