[Bug tree-optimization/104288] [11/12 Regression] EVRP null pointer check removal for strcmp (and maybe others) is not flow senative

["christian.prochaska@genode-labs.com" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104288

Christian Prochaska <christian.prochaska@genode-labs.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |christian.prochaska@genode-
                   |                            |labs.com

--- Comment #13 from Christian Prochaska <christian.prochaska@genode-labs.com> ---
I found the "Register non-null side effects properly." commit with git bisect
while debugging a page fault in the Genode OS framework built with GCC 12.2.0.
It turned out that a null pointer check which was present before this commit is
now not present anymore. The C++ code with the null pointer check can be found
on GitHub:

https://github.com/genodelabs/genode/blob/a84af9a9606450471b8038a35f9b55057efa0850/repos/base-nova/src/lib/base/ipc.cc#L71

This is the implementation of the 'Thread::myself()' function which returns a
null pointer in some conditions:

https://github.com/genodelabs/genode/blob/a84af9a9606450471b8038a35f9b55057efa0850/repos/base/src/lib/base/thread_myself.cc#L22

I compared the disassembled code from objdump and this part is missing when the
commit is applied:

Genode::ipc_call(Genode::Native_capability, Genode::Msgbuf_base&,
Genode::Msgbuf_base&, unsigned long):
/.../repos/base-nova/src/lib/base/ipc.cc:71
    addr_t const manual_rcv_sel = myself ?
myself->native_thread().client_rcv_sel
   85f78:   48 83 bd 50 ff ff ff    cmpq   $0x0,-0xb0(%rbp)
   85f7f:   00
   85f80:   48 c7 c3 ff ff ff ff    mov    $0xffffffffffffffff,%rbx
   85f87:   74 1d                   je     85fa6
<Genode::ipc_call(Genode::Native_capability, Genode::Msgbuf_base&,
Genode::Msgbuf_base&, unsigned long)
/.../repos/base-nova/src/lib/base/ipc.cc:71 (discriminator 1)

Now I'm not sure if the problem is in the Genode code or in GCC. Any ideas?