Re: PING: [PATCH v2] tree-profile: Don't instrument an IFUNC resolver nor its callees

[Jan Hubicka <hubicka@ucw.cz>]

> > I am bit worried about commonly used functions getting "infected" by
> > being called once from ifunc resolver.  I think we only use thread local
> > storage for indirect call profiling, so we may just disable indirect
> > call profiling for these functions.
> 
> Will change it.
> 
> > Also the patch will be noop with -flto -flto-partition=max, so probably
> > we need to compute this flag at WPA time and stream to partitions.
> >
> 
> Why is it a nop with -flto -flto-partition=max? I got
> 
> (gdb) bt
> #0  symtab_node::check_ifunc_callee_symtab_nodes ()
>     at /export/gnu/import/git/gitlab/x86-gcc/gcc/symtab.cc:1440
> #1  0x0000000000e487d3 in symbol_table::compile (this=0x7fffea006000)
>     at /export/gnu/import/git/gitlab/x86-gcc/gcc/cgraphunit.cc:2320
> #2  0x0000000000d23ecf in lto_main ()
>     at /export/gnu/import/git/gitlab/x86-gcc/gcc/lto/lto.cc:687
> #3  0x00000000015254d2 in compile_file ()
>     at /export/gnu/import/git/gitlab/x86-gcc/gcc/toplev.cc:449
> #4  0x00000000015284a4 in do_compile ()
>     at /export/gnu/import/git/gitlab/x86-gcc/gcc/toplev.cc:2154
> #5  0x0000000001528864 in toplev::main (this=0x7fffffffd84a, argc=16,
>     argv=0x42261f0) at /export/gnu/import/git/gitlab/x86-gcc/gcc/toplev.cc:2310
> #6  0x00000000030a3fe2 in main (argc=16, argv=0x7fffffffd958)
>     at /export/gnu/import/git/gitlab/x86-gcc/gcc/main.cc:39
> 
> Do you have a testcase to show that it is a nop?
Aha, sorry.  I tought this is run during late optimization, but it is
done early, so LTo partitioning does not mix things up.  So current
patch modified to disable only instrumentation that needs TLS should be
fine.

Honza
> 
> -- 
> H.J.