Hi! As the following testcase shows, I forgot to call c_fully_fold on the __atomic_*/__sync_* operands called on _BitInt address, the expressions are then used inside of TARGET_EXPR initializers etc. and are never fully folded later, which means we can ICE e.g. on C_MAYBE_CONST_EXPR trees inside of those.
The following patch fixes it, while the function currently is only called in the C FE because C++ doesn't support BITINT_TYPE, I think guarding the calls on !c_dialect_cxx () is safer. Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2024-01-22 Jakub Jelinek <ja...@redhat.com> PR c/113518 * c-common.cc (atomic_bitint_fetch_using_cas_loop): Call c_fully_fold on lhs_addr, val and model for C. * gcc.dg/bitint-77.c: New test. --- gcc/c-family/c-common.cc.jj 2024-01-03 12:07:03.384717147 +0100 +++ gcc/c-family/c-common.cc 2024-01-22 14:24:44.209262552 +0100 @@ -8082,6 +8082,12 @@ atomic_bitint_fetch_using_cas_loop (loca tree lhs_addr = (*orig_params)[0]; tree val = convert (nonatomic_lhs_type, (*orig_params)[1]); tree model = convert (integer_type_node, (*orig_params)[2]); + if (!c_dialect_cxx ()) + { + lhs_addr = c_fully_fold (lhs_addr, false, NULL); + val = c_fully_fold (val, false, NULL); + model = c_fully_fold (model, false, NULL); + } if (TREE_SIDE_EFFECTS (lhs_addr)) { tree var = create_tmp_var_raw (TREE_TYPE (lhs_addr)); --- gcc/testsuite/gcc.dg/bitint-77.c.jj 2024-01-22 14:40:22.505993511 +0100 +++ gcc/testsuite/gcc.dg/bitint-77.c 2024-01-22 14:40:02.445276390 +0100 @@ -0,0 +1,25 @@ +/* PR c/113518 */ +/* { dg-do compile { target bitint } } */ +/* { dg-options "-O2 -std=c23" } */ + +#if __BITINT_MAXWIDTH__ >= 607 +_BitInt(607) v; +#else +_BitInt(63) v; +#endif + +void +foo (void) +{ + __atomic_fetch_or (&v, 1 << 31, __ATOMIC_RELAXED); +} + +#if __BITINT_MAXWIDTH__ >= 16321 +_BitInt(16321) w; + +void +bar (void) +{ + __atomic_fetch_add (&w, 1 << 31, __ATOMIC_SEQ_CST); +} +#endif Jakub