On Wed, 26 Feb 2025, Jakub Jelinek wrote:
> On Wed, Feb 26, 2025 at 10:58:26AM +0100, Richard Biener wrote:
> > > This PR is about ubsan error on the c - cx1 + cy1 evaluation in the first
> > > hunk.
> > >
> > > The following patch hopefully fixes that by doing the
> > > additions/subtractions
> > > in poly_uint64 rather than poly_int64. Or shall we instead perform it in
> > > offset_int and watch for overflows and punt somehow for those?
> >
> > I think when we have the offset computation overflow ignoring such
> > overflow will make the memrefs_conflict_p give possibly wrong
> > answers. In the PR you say cselib now has those
> > -9223372036854775807ish offsets from sp, why does it do alias queries
> > with those clearly invalid offsets?
>
> It wants to have a MEM which overlaps anything below the stack.
> So, uses for stack grows down and non-biased stack sp - PTRDIFF_MAX with
> PTRDIFF_MAX MEM_SIZE as an approximation to that.
I see. Wouldn't setting MEM_OFFSET_KNOWN_P and MEM_SIZE_KNOWN_P
to false work as well?
> > So yes, I think we need to punt on overflow. Using poly_offset_int
> > should work but it comes at a cost. Does poly_wide_int have the same
> > wide-int like overflow overloads?
>
> poly_offset_int will be definitely cheaper than other poly_wide_int, it
> would need to be at least 66 bits and so 128 bit is certainly cheaper then.
>
> So like this if it passes full bootstrap/regtest?
Yes, that looks good to me.
Thanks,
Richard.
> 2025-02-26 Jakub Jelinek <ja...@redhat.com>
>
> PR middle-end/118819
> * alias.cc (memrefs_conflict_p): Perform arithmetics on c, xsize and
> ysize in poly_offset_int and return -1 if it is not representable in
> poly_int64.
>
> --- gcc/alias.cc.jj 2025-01-02 11:23:24.000000000 +0100
> +++ gcc/alias.cc 2025-02-26 12:31:52.860341105 +0100
> @@ -2535,19 +2535,39 @@ memrefs_conflict_p (poly_int64 xsize, rt
> return memrefs_conflict_p (xsize, x1, ysize, y1, c);
> if (poly_int_rtx_p (x1, &cx1))
> {
> + poly_offset_int co = c;
> + co -= cx1;
> if (poly_int_rtx_p (y1, &cy1))
> - return memrefs_conflict_p (xsize, x0, ysize, y0,
> - c - cx1 + cy1);
> + {
> + co += cy1;
> + if (!co.to_shwi (&c))
> + return -1;
> + return memrefs_conflict_p (xsize, x0, ysize, y0, c);
> + }
> + else if (!co.to_shwi (&c))
> + return -1;
> else
> - return memrefs_conflict_p (xsize, x0, ysize, y, c - cx1);
> + return memrefs_conflict_p (xsize, x0, ysize, y, c);
> }
> else if (poly_int_rtx_p (y1, &cy1))
> - return memrefs_conflict_p (xsize, x, ysize, y0, c + cy1);
> + {
> + poly_offset_int co = c;
> + co += cy1;
> + if (!co.to_shwi (&c))
> + return -1;
> + return memrefs_conflict_p (xsize, x, ysize, y0, c);
> + }
>
> return -1;
> }
> else if (poly_int_rtx_p (x1, &cx1))
> - return memrefs_conflict_p (xsize, x0, ysize, y, c - cx1);
> + {
> + poly_offset_int co = c;
> + co -= cx1;
> + if (!co.to_shwi (&c))
> + return -1;
> + return memrefs_conflict_p (xsize, x0, ysize, y, c);
> + }
> }
> else if (GET_CODE (y) == PLUS)
> {
> @@ -2563,7 +2583,13 @@ memrefs_conflict_p (poly_int64 xsize, rt
>
> poly_int64 cy1;
> if (poly_int_rtx_p (y1, &cy1))
> - return memrefs_conflict_p (xsize, x, ysize, y0, c + cy1);
> + {
> + poly_offset_int co = c;
> + co += cy1;
> + if (!co.to_shwi (&c))
> + return -1;
> + return memrefs_conflict_p (xsize, x, ysize, y0, c);
> + }
> else
> return -1;
> }
> @@ -2616,8 +2642,16 @@ memrefs_conflict_p (poly_int64 xsize, rt
> if (maybe_gt (xsize, 0))
> xsize = -xsize;
> if (maybe_ne (xsize, 0))
> - xsize += sc + 1;
> - c -= sc + 1;
> + {
> + poly_offset_int xsizeo = xsize;
> + xsizeo += sc + 1;
> + if (!xsizeo.to_shwi (&xsize))
> + return -1;
> + }
> + poly_offset_int co = c;
> + co -= sc + 1;
> + if (!co.to_shwi (&c))
> + return -1;
> return memrefs_conflict_p (xsize, canon_rtx (XEXP (x, 0)),
> ysize, y, c);
> }
> @@ -2631,8 +2665,16 @@ memrefs_conflict_p (poly_int64 xsize, rt
> if (maybe_gt (ysize, 0))
> ysize = -ysize;
> if (maybe_ne (ysize, 0))
> - ysize += sc + 1;
> - c += sc + 1;
> + {
> + poly_offset_int ysizeo = ysize;
> + ysizeo += sc + 1;
> + if (!ysizeo.to_shwi (&ysize))
> + return -1;
> + }
> + poly_offset_int co = c;
> + co += sc + 1;
> + if (!co.to_shwi (&c))
> + return -1;
> return memrefs_conflict_p (xsize, x,
> ysize, canon_rtx (XEXP (y, 0)), c);
> }
> @@ -2643,7 +2685,11 @@ memrefs_conflict_p (poly_int64 xsize, rt
> poly_int64 cx, cy;
> if (poly_int_rtx_p (x, &cx) && poly_int_rtx_p (y, &cy))
> {
> - c += cy - cx;
> + poly_offset_int co = c;
> + co += cy;
> + co -= cx;
> + if (!co.to_shwi (&c))
> + return -1;
> return offset_overlap_p (c, xsize, ysize);
> }
>
>
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)
