Re: [PATCH, middle-end]: Fix PR68999, gfortran.fortran-torture/execute/save_1.f90 execution failure on alpha

[Uros Bizjak <ubizjak at gmail dot com>]

On Wed, Dec 23, 2015 at 7:30 PM, Richard Biener
<richard.guenther@gmail.com> wrote:
> On December 23, 2015 5:58:07 PM GMT+01:00, Uros Bizjak <ubizjak@gmail.com> wrote:
>>On Wed, Dec 23, 2015 at 2:39 PM, Richard Biener
>><richard.guenther@gmail.com> wrote:
>>> On December 23, 2015 10:39:17 AM GMT+01:00, Uros Bizjak
>><ubizjak@gmail.com> wrote:
>>>>Hello!
>>>>
>>>>There is a logic error in Honza's patch "Transparent alias suport
>>part
>>>>10" [1]. The part in memrefs_conflict_p should be changed to:
>>>>
>>>>-      /* If decls are different or we know by offsets that there is
>>no
>>>>overlap,
>>>>- we win.  */
>>>>-      if (!cmp || !offset_overlap_p (c, xsize, ysize))
>>>>+      /* If decls are different and we know by offsets that
>>>>+ there is no overlap, we win.  */
>>>>+      if (!cmp && !offset_overlap_p (c, xsize, ysize))
>>>>  return 0;
>>>>-      /* Decls may or may not be different and offsets overlap....*/
>>>>+      /* Decls are different and offsets overlap....*/
>>>>
>>>>Even if decls are different, their offsets shouldn't overlap!
>>>
>>> Comparing offsets of different decls does not make sense.
>>
>>Uh, yes, some more eyeballing was needed, but you are right.
>>
>>Is there a way to detect aliasing in case AND addresses are involved?
>>
>>Probably we need something like in base_alias_check, where:
>
> Yeah, and in that case just give up.

As mentioned in [1], in attached v2 patch, we return "unknown" from
memrefs_conflict_p when realigned decls are processed:

@@ -2339,6 +2337,12 @@ memrefs_conflict_p (int xsize, rtx x, int ysize, r
       /* If both decls are the same, decide by offsets.  */
       if (cmp == 1)
         return offset_overlap_p (c, xsize, ysize);
+      /* Assume a potential overlap for symbolic addresses that went
+        through alignment adjustments (i.e., that have negative
+        sizes), because we can't know how far they are from each
+        other.  */
+      if (xsize < 0 || ysize < 0)
+       return -1;
       /* If decls are different or we know by offsets that there is no overlap,
         we win.  */
       if (!cmp || !offset_overlap_p (c, xsize, ysize))

This is the same approach as it is done at the end of memrefs_conflict_p.

We still need early return for AND addresses in base_alias_check, though.

2015-12-29  Uros Bizjak  <ubizjak@gmail.com>

    PR middle-end/68999
    * symtab.c (symtab_node::equal_address_to): Return -1 instead of 2
    if we can't determine address equivalence.
    * alias.c (compare_base_decl): Update for changed return value of
    symtab_node::equal_address_to.
    (memrefs_conflict_p): Return -1 for different decls that went through
    alignment adjustments.
    (base_alias_check): Move check for addresses with alignment ANDs
    before the call for compare_base_decls.

Patch was bootstrapped and regression tested on x86_64-linux-gnu
{,-m32} and alpha-linux-gnu native [2].

OK for mainline?

[1] https://gcc.gnu.org/ml/gcc-patches/2015-12/msg02174.html
[2] https://gcc.gnu.org/ml/gcc-testresults/2015-12/msg02766.html

Uros.

Attachment: p.diff.txt
Description: Text document