https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109231
--- Comment #30 from Jakub Jelinek <jakub at gcc dot gnu.org> --- So, if I add the --- a/gcc/tree-inline.cc +++ b/gcc/tree-inline.cc @@ -2787,6 +2787,7 @@ initialize_cfun (tree new_fndecl, tree callee_fndecl, profile_count count) /* Get clean struct function. */ push_struct_function (new_fndecl, true); targetm.target_option.relayout_function (new_fndecl); + aggregate_value_p (DECL_RESULT (new_fndecl), new_fndecl); /* We will rebuild these, so just sanity check that they are empty. */ gcc_assert (VALUE_HISTOGRAMS (cfun) == NULL); I see 5 calls of this in this spot, two with int return type, one with TFmode floating one, one with uint and one with bool. aggregate_value_p calls targetm.calls.return_in_memory but I don't see sparc_return_in_memory to have any side-effects other than return value. Next it calls reg = hard_function_value (type, 0, fntype, 0); which does allocate a REG rtx, but doesn't save it anywhere. It then calls fntype_abi and hard_regno_nregs and predefined_function_abi::clobbers_full_reg_p, but I don't see any side-effects in those either. Nothing seems to cache its results in any way. Confused...
