clang optimization issue in coroutines(not allocations)

[kelbon]

https://godbolt.org/z/5hTE69j1b

starting row 82 in asm there are something strange in 'trunk', im 99% sure someone apply 'cold' attribute or 'never inline' to all functions in final suspend or smth like.

All calls must be inlined, its literaly one-row functions.

g1() [clone .resume]:                          # @g1() [clone .resume]
        push    rbx
        sub     rsp, 16
        cmp     byte ptr [rdi + 44], 0
        je      .LBB8_1
        lea     rbx, [rdi + 16]
        mov     qword ptr [rdi], 0
        call    std::__n4861::coroutine_handle<dd::generator_promise<int> >::from_address(void*)
        mov     qword ptr [rsp + 8], rax
        lea     rdi, [rsp + 8]
        call    std::__n4861::coroutine_handle<dd::generator_promise<int> >::operator std::__n4861::coroutine_handle<void>() const
        mov     rdi, rbx
        mov     rsi, rax
        call    dd::generator_promise<int>::await_suspend(std::__n4861::coroutine_handle<void>) const
        mov     qword ptr [rsp], rax
        mov     rdi, rsp
        call    std::__n4861::coroutine_handle<void>::address() const
        mov     rdi, rax
        add     rsp, 16
        pop     rbx
        jmp     qword ptr [rax]                 # TAILCALL
.LBB8_1:

[llvmbot]

@llvm/issue-subscribers-coroutines

[ChuanqiXu9]

Sorry this is a known regression after c467245. And the proposed solution may solve the issue: #64945. Then let's track the issue there.