https://gcc.gnu.org/bugzilla/show_bug.cgi?id=43798
--- Comment #12 from rguenther at suse dot de <rguenther at suse dot de> --- On Wed, 20 Mar 2019, jakub at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=43798 > > --- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> --- > We can't pad, that is against the ABI. > I think we've discussed this many times in the past, the fact is that people > do > use this kind of mess when they want to say that the whole array has certain > alignment or when they are just careless, I think it is too late now for GCC9 > to start rejecting it without an analysis how much code would be affected, not > to mention that it is not backportable. > > The following patch does the adjustment and fixes the testcase: > --- gcc/stor-layout.c.jj 2019-01-01 12:37:17.296972670 +0100 > +++ gcc/stor-layout.c 2019-03-20 09:56:15.576407411 +0100 > @@ -2550,10 +2550,27 @@ layout_type (tree type) > /* If TYPE_SIZE_UNIT overflowed, then it is certainly larger than > TYPE_ALIGN_UNIT. */ > && !TREE_OVERFLOW (TYPE_SIZE_UNIT (element)) > - && !integer_zerop (TYPE_SIZE_UNIT (element)) > - && compare_tree_int (TYPE_SIZE_UNIT (element), > - TYPE_ALIGN_UNIT (element)) < 0) I believe this check was just bogus since size < align catches just very few cases of small objects. Thus the case in question now isn't really different from the intent of the above. Shouldn't we just change this to && !multiple_of_p (sizetype, TYPE_SIZE_UNIT (element), TYPE_ALIGN_UNIT (element)) and adjust the error message accordingly? > - error ("alignment of array elements is greater than element size"); > + && !integer_zerop (TYPE_SIZE_UNIT (element))) > + { > + if (compare_tree_int (TYPE_SIZE_UNIT (element), > + TYPE_ALIGN_UNIT (element)) < 0) > + error ("alignment of array elements is greater than " > + "element size"); > + if (TYPE_ALIGN_UNIT (element)) > + { > + unsigned align = (TREE_INT_CST_LOW (TYPE_SIZE_UNIT (element)) > + & (TYPE_ALIGN_UNIT (element) - 1)); > + align = least_bit_hwi (align); > + if (align && align != TYPE_ALIGN_UNIT (element)) > + { > + element = build_variant_type_copy (element); > + SET_TYPE_ALIGN (element, align * BITS_PER_UNIT); > + TYPE_USER_ALIGN (element) = 1; > + TREE_TYPE (type) = element; > + TYPE_USER_ALIGN (type) = 1; > + } > + } > + } > break; > } > > but I haven't done any testing on it beyond that. > Perhaps for GCC 9 we could warn when we do this and depending on the amount of > warnings consider rejecting it later on if it doesn't affect significant > amount > of code? > >
