On Tue, Aug 05, 2025 at 12:04:56PM GMT, Jakub Jelinek wrote:
> On Sat, Aug 02, 2025 at 05:14:22PM +0800, Yang Yujie wrote:
> > tree
> > -bitint_large_huge::limb_access (tree type, tree var, tree idx, bool
> > write_p)
> > +bitint_large_huge::limb_access (tree type, tree var, tree idx, bool
> > write_p,
> > + bool abi_load_p)
> > {
> > tree atype = (tree_fits_uhwi_p (idx)
> > ? limb_access_type (type, idx) : m_limb_type);
> > - tree ltype = m_limb_type;
> > +
> > + tree ltype = (bitint_extended && abi_load_p) ? atype : m_limb_type;
> > +
> > addr_space_t as = TYPE_ADDR_SPACE (TREE_TYPE (var));
> > if (as != TYPE_ADDR_SPACE (ltype))
> > ltype = build_qualified_type (ltype, TYPE_QUALS (ltype)
> > @@ -651,12 +654,21 @@ bitint_large_huge::limb_access (tree type, tree var,
> > tree idx, bool write_p)
> > {
> > unsigned HOST_WIDE_INT nelts
> > = CEIL (tree_to_uhwi (TYPE_SIZE (TREE_TYPE (var))), limb_prec);
> > - tree atype = build_array_type_nelts (ltype, nelts);
> > +
> > + /* Build the array type with m_limb_type from the right address
> > + space. */
> > + tree limb_type_a = m_limb_type;
> > + if (as != TYPE_ADDR_SPACE (m_limb_type))
> > + limb_type_a = build_qualified_type (m_limb_type,
> > + TYPE_QUALS (m_limb_type)
> > + | ENCODE_QUAL_ADDR_SPACE (as));
> > +
> > + tree atype = build_array_type_nelts (limb_type_a, nelts);
> > var = build1 (VIEW_CONVERT_EXPR, atype, var);
> > }
> > ret = build4 (ARRAY_REF, ltype, var, idx, NULL_TREE, NULL_TREE);
> > }
> > - if (!write_p && !useless_type_conversion_p (atype, m_limb_type))
> > + if (!write_p && !useless_type_conversion_p (atype, ltype))
>
> This part is not correct. For non-generic address space this
> will be always non-useless conversion. If you want to do nothing
> for the bitint_extended && abi_load_p case, I'd write it as
> if (!write_p
> && !(bitint_extended && abi_load_p)
> && !useless_type_conversion_p (atype, m_limb_type))
>
> Jakub
I am not familiar with the non-generic address spaces, but I assume it would be
OK to have a final cast on the result as long as it is not a truncation,
which cannot be optimized away later.
Also, this cast seems to serve a purpose if loading from these address
spaces is not a no-op (not sure about this too).
Yujie
