https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121692
--- Comment #4 from jchrist at linux dot ibm.com --- (In reply to Andrew Pinski from comment #3) > So with the cold attribute, small "functions" don't inline always: > ``` > Deciding on inlining of small functions. Starting with size 0. > > /app/example.cpp:66:15: missed: not inlinable: __order_base_2.part.0/17 -> > __flogr/1, call is unlikely and code size would grow > > ``` That is exactly the problem. Since all the __builtin_constant_p paths through the function are folded away later (and, especially, at the call sites, the input is not compile-time constant anymore), the function size should actually be small. Nevertheless, size computation includes all the statements of the __builtin_constant_p branches. Our kernel devs have such patterns a lot in the code and assume that the code gets properly inlined. Nevertheless, we see a lot of out-of-line functions based on this. > Maybe cold is not the right attribute here. Well, that comes from the kernel init functions and at least semantically is right since the init function is called only once. I see inlining into non-cold functions, but as soon as only cold functions are left to inline into, we get an out-of-line function.
