On Fri, Jul 25, 2025 at 10:53:37AM +0800, Yang Yujie wrote:
> This patch also make casts of ABI-extended large/huge _BitInts
> behave the same as the small/middle case, i.e. no-op for casts
> to a higher precision _BitInt with the same number of limbs /
> extension for casts that turns a full top limb into a partial limb.
> This conveniently helps keep some code with implementation-specific
> extension semantics (e.g. BEXTC from gcc.dg/bitintext.h) the same
> for _BitInts of any precision.
>
> gcc/ChangeLog:
>
> * gimple-lower-bitint.cc (bitint_large_huge::limb_access):
> Add a parameter abi_load_p. If set, load a limb directly
> in its actual precision without casting from m_limb_type.
> (struct bitint_large_huge): Same.
> (bitint_large_huge::handle_load): Use.
> ---
> gcc/gimple-lower-bitint.cc | 27 +++++++++++++++++++--------
> 1 file changed, 19 insertions(+), 8 deletions(-)
>
> diff --git a/gcc/gimple-lower-bitint.cc b/gcc/gimple-lower-bitint.cc
> index 8fb7a604591..631a7844790 100644
> --- a/gcc/gimple-lower-bitint.cc
> +++ b/gcc/gimple-lower-bitint.cc
> @@ -429,7 +429,7 @@ struct bitint_large_huge
>
> void insert_before (gimple *);
> tree limb_access_type (tree, tree);
> - tree limb_access (tree, tree, tree, bool);
> + tree limb_access (tree, tree, tree, bool, bool = false);
> tree build_bit_field_ref (tree, tree, unsigned HOST_WIDE_INT,
> unsigned HOST_WIDE_INT);
> void if_then (gimple *, profile_probability, edge &, edge &);
> @@ -610,11 +610,15 @@ bitint_large_huge::limb_access_type (tree type, tree
> idx)
> TYPE. If WRITE_P is true, it will be a store, otherwise a read. */
>
> 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 = (tree_fits_uhwi_p (idx) && bitint_extended && abi_load_p
> + ? limb_access_type (type, idx) : m_limb_type);
Why do you call limb_access_type again? atype in that case
should be the result of limb_access_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 +655,12 @@ 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);
> + tree atype = build_array_type_nelts (m_limb_type, nelts);
This looks wrong. It needs to be from the right address space, see above
if (as != TYPE_ADDR_SPACE (ltype))
ltype = build_qualified_type ...
> 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))
Again, this looks wrong for non-standard address spaces.
Jakub
