Hi! The bitint_extended changes in limb_access broke bitint_big_endian. As we sometimes (for bitint_extended) access the MEM_REFs using atype rather than m_limb_type, for big-endian we need to adjust the MEM_REFs offset if atype has smaller TYPE_SIZE_UNIT than m_limb_size.
Fixed thusly, bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk if Stefan's testing on s390x-linux succeeds? 2025-11-27 Jakub Jelinek <[email protected]> PR target/122714 * gimple-lower-bitint.cc (bitint_large_huge::limb_access): Adjust MEM_REFs offset for bitint_big_endian if ltype doesn't have the same byte size as m_limb_type. --- gcc/gimple-lower-bitint.cc.jj 2025-09-30 13:37:48.364854566 +0200 +++ gcc/gimple-lower-bitint.cc 2025-11-26 17:03:51.815649841 +0100 @@ -630,6 +630,8 @@ bitint_large_huge::limb_access (tree typ | ENCODE_QUAL_ADDR_SPACE (as)); tree ptype = build_pointer_type (strip_array_types (TREE_TYPE (var))); unsigned HOST_WIDE_INT off = tree_to_uhwi (idx) * m_limb_size; + if (bitint_big_endian) + off += m_limb_size - tree_to_uhwi (TYPE_SIZE_UNIT (ltype)); ret = build2 (MEM_REF, ltype, build_fold_addr_expr (var), build_int_cst (ptype, off)); @@ -641,12 +643,14 @@ bitint_large_huge::limb_access (tree typ if (as != TYPE_ADDR_SPACE (ltype)) ltype = build_qualified_type (ltype, TYPE_QUALS (ltype) | ENCODE_QUAL_ADDR_SPACE (as)); + unsigned HOST_WIDE_INT off = tree_to_uhwi (idx) * m_limb_size; + if (bitint_big_endian) + off += m_limb_size - tree_to_uhwi (TYPE_SIZE_UNIT (ltype)); ret = build2 (MEM_REF, ltype, unshare_expr (TREE_OPERAND (var, 0)), size_binop (PLUS_EXPR, TREE_OPERAND (var, 1), build_int_cst (TREE_TYPE (TREE_OPERAND (var, 1)), - tree_to_uhwi (idx) - * m_limb_size))); + off))); TREE_THIS_VOLATILE (ret) = TREE_THIS_VOLATILE (var); TREE_SIDE_EFFECTS (ret) = TREE_SIDE_EFFECTS (var); TREE_THIS_NOTRAP (ret) = TREE_THIS_NOTRAP (var); Jakub
