lftr and str are inherently unsafe optimiztion. your change might help
a little, but won't make the opts safe. I suggest you run more test to
ensure no regression
Sun

On Wed, Feb 23, 2011 at 2:58 AM, Ye, Mei <mei...@amd.com> wrote:
> For 32-bit targets, when address of data reference crosses the boundary of
> 0x80000000, signed comparisons of address expressions will lead to
>  incorrect results.
>
> Additional comments and patches are also available in the bug database.
>
>
>
> My fix is:
>
> 1. In LFTR, force comparison of address expressions to use unsigned type for
>
> 32-bit targets.
>
> 2. Allow results of pointer arithmetic "p + a" and "p - a" (where "p" is a
>
> pointer or an address expression, "a" is neither a pointer nor an address
>
> expression") to be recognized as an address expression.
>
>
>
> Index: be/opt/opt_htable.cxx
>
> ===================================================================
>
> --- be/opt/opt_htable.cxx           (revision 3488)
>
> +++ be/opt/opt_htable.cxx        (working copy)
>
> @@ -5102,8 +5102,11 @@
>
>        if (res0 == ADDRESSABILITY_UNKNOWN)
>
>                return ADDRESSABILITY_UNKNOWN;
>
>        res1 = Opnd(1)->Check_if_result_is_address(opt_stab);
>
> -      if (res1 == ADDRESSABILITY_UNKNOWN)
>
> +      if (res1 == ADDRESSABILITY_UNKNOWN) {
>
> +             if (res0 == ADDRESSABILITY_IS_ADDRESS)
>
> +               return res0;
>
>                return ADDRESSABILITY_UNKNOWN;
>
> +      }
>
>        if (res0 == res1) {
>
>          if (res0 == ADDRESSABILITY_IS_ADDRESS)
>
>            return ADDRESSABILITY_NOT_ADDRESS;
>
> Index: be/opt/opt_lftr2.cxx
>
> ===================================================================
>
> --- be/opt/opt_lftr2.cxx                (revision 3488)
>
> +++ be/opt/opt_lftr2.cxx             (working copy)
>
> @@ -1216,11 +1216,18 @@
>
>      // adjust operator of new comparison
>
>      OPERATOR new_compare_opr = comparison_cr->Opr();
>
> -#ifdef KEY
>
> +#ifdef KEY
>
>      MTYPE new_compare_type = tempcr->Dtyp();
>
>      // do not change signedness of comparison since that could change
> semantics
>
>      new_compare_type = Mtype_TransferSign(comparison_cr->Dsctyp(),
>
>
>   new_compare_type);
>
> +    if (Is_Target_32bit()
>
> +             && (addressable == ADDRESSABILITY_IS_ADDRESS)
>
> +             && MTYPE_is_signed(new_compare_type)) {
>
> +      // For 32-bit targets, force comparison of address expressions to be
> unsigned.
>
> +      new_compare_type =
> Mtype_from_mtype_class_and_size(MTYPE_CLASS_UNSIGNED,
>
> +
> MTYPE_size_min(new_compare_type)/8);
>
> +    }
>
> #else
>
>      MTYPE new_compare_type = comparison_cr->Dsctyp();
>
>      if (addressable == ADDRESSABILITY_IS_ADDRESS &&
>
> ------------------------------------------------------------------------------
> Free Software Download: Index, Search & Analyze Logs and other IT data in
> Real-Time with Splunk. Collect, index and harness all the fast moving IT
> data
> generated by your applications, servers and devices whether physical,
> virtual
> or in the cloud. Deliver compliance at lower cost and gain new business
> insights. http://p.sf.net/sfu/splunk-dev2dev
> _______________________________________________
> Open64-devel mailing list
> Open64-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/open64-devel
>
>

------------------------------------------------------------------------------
Free Software Download: Index, Search & Analyze Logs and other IT data in 
Real-Time with Splunk. Collect, index and harness all the fast moving IT data 
generated by your applications, servers and devices whether physical, virtual
or in the cloud. Deliver compliance at lower cost and gain new business 
insights. http://p.sf.net/sfu/splunk-dev2dev 
_______________________________________________
Open64-devel mailing list
Open64-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/open64-devel

Reply via email to