exchange_and_add inlined in g++ 4.0.2

[bert hubert]

On Sun, Nov 06, 2005 at 02:08:53AM +0100, Paolo Carlini wrote:
> This problem dates back to february, 2004:
>     http://gcc.gnu.org/ml/libstdc++/2004-02/msg00372.html

Thanks for the reference. So do I understand correctly there is no way to
get the i486+ version?

> and we really hoped the performance would not be measurably affected...
> You say "appear in the profile", indeed, I can confirm that recently I
> saw it myself quite a few times, but I wonder whether it's very high in
> your profile or not, can you quantify?

It is the top function that is related to the pdns_recursor in this system-wide profile:
CPU: Pentium M (P6 core), speed 1398.97 MHz (estimated)
Counted CPU_CLK_UNHALTED events (clocks processor is not halted, and not in a thermal trip) with a unit mask of 0x00 (No unit mask) count 100000
samples  %        image name               app name                 symbol name
366472   48.6481  vmlinux-2.6.14-ahu       vmlinux-2.6.14-ahu       poll_idle
31923     4.2377  anon (tgid:4494 range:0x81fd000-0x86ab000) Xorg                     (no symbols)
15241     2.0232  libstdc++.so.6.0.5       libstdc++.so.6.0.5       __gnu_cxx::__exchange_and_add(int volatile*, int)
11936     1.5845  vmlinux-2.6.14-ahu       vmlinux-2.6.14-ahu       get_offset_pmtmr
10314     1.3692  libstdc++.so.6.0.5       libstdc++.so.6.0.5       std::basic_string<char, std::char_traits<char>, std::allocator<char> >::basic_string(std::string const&)
10160     1.3487  libc-2.3.5.so            libc-2.3.5.so            vfprintf
9760      1.2956  libstdc++.so.6.0.5       libstdc++.so.6.0.5       __gnu_cxx::__mt_alloc<char, __gnu_cxx::__common_pool_policy<__gnu_cxx::__pool, true> >::allocate(unsigned int,
void const*)
9154      1.2152  libstdc++.so.6.0.5       libstdc++.so.6.0.5       __gnu_cxx::__pool<true>::_M_reclaim_block(char*, unsigned int)
8759      1.1627  libc-2.3.5.so            libc-2.3.5.so            memcpy
8315      1.1038  libstdc++.so.6.0.5       libstdc++.so.6.0.5       __gnu_cxx::__atomic_add(int volatile*, int)
7369      0.9782  libstdc++.so.6.0.5       libstdc++.so.6.0.5       std::basic_string<char, std::char_traits<char>, std::allocator<char> >::~basic_string()
6341      0.8417  libstdc++.so.6.0.5       libstdc++.so.6.0.5       .plt
5996      0.7960  libc-2.3.5.so            libc-2.3.5.so            _int_malloc
5203      0.6907  libc-2.3.5.so            libc-2.3.5.so            memmove
4253      0.5646  pdns_recursor            pdns_recursor            std::_Rb_tree<DNSResourceRecord, DNSResourceRecord, std::_Identity<DNSResourceRecord>, std::less<DNSResourceRecord>, std::allocator<DNSResourceRecord> >::insert_unique(DNSResourceRecord const&)
4040      0.5363  pdns_recursor            pdns_recursor            MOADNSParser::init(char const*, unsigned int)
4032      0.5352  libstdc++.so.6.0.5       libstdc++.so.6.0.5       std::string::_Rep::_S_create(unsigned int, unsigned int, std::allocator<char> const&)
3834      0.5090  pdns_recursor            pdns_recursor            MemRecursorCache::doPrune()
3830      0.5084  libstdc++.so.6.0.5       libstdc++.so.6.0.5       __gnu_cxx::__pool<true>::_M_get_thread_id()

> This is a very reasonable request, in my opinion. Actually, I considered
> briefly that issue some time ago, then got distracted, unfortunately.
> Can you possibly file a complete Bugzilla PR describing this enhancement
> request? You can also put there, immediately, or afterwards, any
> additional info as per the point above.

Will do. I think I can provide my own __exchange_and_add as an inline?

Thanks!

-- 
http://www.PowerDNS.com      Open source, database driven DNS Software 
http://netherlabs.nl              Open and Closed source services