https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118945
--- Comment #3 from Vineet Gupta <vineetg at gcc dot gnu.org> --- (In reply to JuzheZhong from comment #2) > I have thought about this long time ago while I am working on supporting RVV > on upstream GCC. > > https://github.com/riscv-non-isa/riscv-toolchain-conventions/issues/37 > > I suggested we should have -mprefer-agnostic in RISC-V psabi. > > When this option is allowed, the problem can be solved very easily: > > > /* Get prefer tail policy. */ > enum tail_policy > get_prefer_tail_policy () > { > /* TODO: By default, we choose to use TAIL_ANY which allows > compiler pick up either agnostic or undisturbed. Maybe we > will have a compile option like -mprefer=agnostic to set > this value???. */ > return TAIL_ANY; > } > ... > in riscv-v.cc. I have already leave a TODO here. Yeah I saw that TODO already and unfortunately just forcing the above function to return TAIL_AGNOSTIC isn't enough to stop merging - for this test (whether we use -mprefer-agnostic or a cpu tune is just an implementation detail and the easy part) It seems we need to change policy_compatible_p () which currently does the following (in macro expanded code) if (prev_flags == policy_demand_type::ignore_policy && next_flags == policy_demand_type::tail_policy_only) return always_true (prev, next); Since we have insn 70 with ignore_policy, it doesn't really check the m_ta for both the infos.
