Performance Discrepancy Observed in Self-Built Version

[FlameTornado10]

Hello Charles,

First I want to thank you for the great work on this spsc queue project. I recently compiled and ran all fifos , but I've noticed that the performance of the build does not match the benchmarks or performance metrics stated in the cppcon2023 video. Here are the details of my environment and the results I observed:

My Environment

• Operating System: WSL2 Ubuntu 22.04.5
• Processor: 13th Gen Intel(R) Core(TM) i9-13900H 24 MiB L3 Cache
• Compiler/Toolchain: g++-12

My expected performance:
As in the video, fifo4 is 10 times that of fifo2, and fifo4a is 4 times that of fifo4.
Actual performance on my WSL:

Fifo2  :    8759994 ops/s
Fifo3  :    9589287 ops/s
Fifo4  :    8774743 ops/s
Fifo4a :    9136957 ops/s
Fifo4b :   10618581 ops/s
Fifo5  :   11817761 ops/s
Fifo5a :   11524272 ops/s
Fifo5b :    9317922 ops/s
rigtorp:    8892872 ops/s
boost::spsc_queue fixed:    8596137 ops/s
Mutex  :    5885189 ops/s

I build without changing the project. Could you please help me understand if I am missing something, or if there are known issues or additional configurations that could impact performance in this way?

[CharlesFrasch]

Hi, I’m on vacation right now and don’t have a computer with me.Make sure you pin the threads.I ran the programs 100 times and took the minimum. There is a lot of jitter in the timesR/cffSent from my iPhoneOn Oct 17, 2024, at 7:26 PM, FlameTornado10 ***@***.***> wrote: Hello Charles, First I want to thank you for the great work on this spsc queue project. I recently compiled and ran all fifos , but I've noticed that the performance of the build does not match the benchmarks or performance metrics stated in the cppcon2023 video. Here are the details of my environment and the results I observed: My Environment Operating System: WSL2 Ubuntu 22.04.5 Processor: 13th Gen Intel(R) Core(TM) i9-13900H 24 MiB L3 Cache Compiler/Toolchain: g++-12 My expected performance: As in the video, fifo4 is 10 times that of fifo2, and fifo4a is 4 times that of fifo4. Actual performance on my WSL: Fifo2 : 8759994 ops/s Fifo3 : 9589287 ops/s Fifo4 : 8774743 ops/s Fifo4a : 9136957 ops/s Fifo4b : 10618581 ops/s Fifo5 : 11817761 ops/s Fifo5a : 11524272 ops/s Fifo5b : 9317922 ops/s rigtorp: 8892872 ops/s boost::spsc_queue fixed: 8596137 ops/s Mutex : 5885189 ops/s I build without changing the project. Could you please help me understand if I am missing something, or if there are known issues or additional configurations that could impact performance in this way? —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[FlameTornado10]

Thanks for your reply! I reran the benchmark 10 times on a Linux server and observed the expected performance: fifo4's performance is 10 times that of fifo2, and fifo4a's performance is 4 times that of fifo4.
But I found another potential problem:
there is an error in this compile command:

/usr/bin/g++-12   -O3 -DNDEBUG -Wall -Werror -Wextra -Wconversion -Wno-unused-parameter -std=c++23 -o CMakeFiles/fifo5.dir/fifo5.cpp.o -c /home/ddzzkk/PRAC/Test/cppcon2023/fifo5.cpp

which is generated by this cmake command:

cd build && cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE
_EXPORT_COMPILE_COMMANDS=ON ../

the error is like this

It says:

ddzzkk@LAPTOP-Xin ~/P/T/c/build (main) [2]> /usr/bin/g++-12   -O3 -DNDEBUG -Wall -We
rror -Wextra -Wconversion -Wno-unused-parameter -std=c++23 -o CMakeFiles/fifo5.dir/f
ifo5.cpp.o -c /home/ddzzkk/PRAC/Test/cppcon2023/fifo5.cpp
In file included from /usr/include/c++/12/atomic:41,
                 from /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:3,
                 from /home/ddzzkk/PRAC/Test/cppcon2023/fifo5.cpp:1:
In member function ‘std::__atomic_base<_IntTp>::__int_type std::__atomic_base<_IntTp>::load(std::memory_order) const [with _ITp = long unsigned int]’,
    inlined from ‘auto Fifo5<T, Alloc>::pop() [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:207:49,
    inlined from ‘auto Fifo5<T, Alloc>::pop(T&) [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:218:18,
    inlined from ‘void Bench<T>::pop(value_type) [with T = Fifo5<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:90:42,
    inlined from ‘Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:52:20,
    inlined from ‘constexpr _Res std::__invoke_impl(__invoke_other, _Fn&&, _Args&& ...) [with _Res = void; _Fn = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:61:36,
    inlined from ‘constexpr typename std::__invoke_result<_Functor, _ArgTypes>::type std::__invoke(_Callable&&, _Args&& ...) [with _Callable = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:96:40,
    inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::_M_invoke(std::_Index_tuple<_Ind ...>) [with long unsigned int ..._Ind = {0}; _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:279:26,
    inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::operator()() [with _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:286:20,
    inlined from ‘void std::thread::_State_impl<_Callable>::_M_run() [with _Callable = std::thread::_Invoker<std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> > >]’ at /usr/include/c++/12/bits/std_thread.h:231:20:
/usr/include/c++/12/bits/atomic_base.h:488:31: error: ‘long unsigned int __atomic_load_8(const volatile void*, int)’ writing 8 bytes into a region of size 0 overflows the destination [-Werror=stringop-overflow=]
  488 |         return __atomic_load_n(&_M_i, int(__m));
      |                ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~
In member function ‘std::__atomic_base<_IntTp>::__int_type std::__atomic_base<_IntTp>::load(std::memory_order) const [with _ITp = long unsigned int]’,
    inlined from ‘auto Fifo5<T, Alloc>::pop() [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:207:49,
    inlined from ‘auto Fifo5<T, Alloc>::pop(T&) [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:218:18,
    inlined from ‘void Bench<T>::pop(value_type) [with T = Fifo5<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:90:42,
    inlined from ‘Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:57:20,
    inlined from ‘constexpr _Res std::__invoke_impl(__invoke_other, _Fn&&, _Args&& ...) [with _Res = void; _Fn = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:61:36,
    inlined from ‘constexpr typename std::__invoke_result<_Functor, _ArgTypes>::type std::__invoke(_Callable&&, _Args&& ...) [with _Callable = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:96:40,
    inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::_M_invoke(std::_Index_tuple<_Ind ...>) [with long unsigned int ..._Ind = {0}; _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:279:26,
    inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::operator()() [with _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:286:20,
    inlined from ‘void std::thread::_State_impl<_Callable>::_M_run() [with _Callable = std::thread::_Invoker<std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> > >]’ at /usr/include/c++/12/bits/std_thread.h:231:20:
/usr/include/c++/12/bits/atomic_base.h:488:31: error: ‘long unsigned int __atomic_load_8(const volatile void*, int)’ writing 8 bytes into a region of size 0 overflows the destination [-Werror=stringop-overflow=]
  488 |         return __atomic_load_n(&_M_i, int(__m));
      |                ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~
cc1plus: all warnings being treated as errors

I search for this error and don't know why it comes out. I debuged the code and found the problem seems to come from the usage of popper_t and its destructor, but I didn't figure out how to fix this.

[CharlesFrasch]

Will be back in US Monday. Will take a lookSent from my iPhoneOn Oct 26, 2024, at 4:49 AM, FlameTornado10 ***@***.***> wrote: Thanks for your reply! I reran the benchmark 10 times on a Linux server and observed the expected performance: fifo4's performance is 10 times that of fifo2, and fifo4a's performance is 4 times that of fifo4. But I found another potential problem: there is an error in this compile command: /usr/bin/g++-12 -O3 -DNDEBUG -Wall -Werror -Wextra -Wconversion -Wno-unused-parameter -std=c++23 -o CMakeFiles/fifo5.dir/fifo5.cpp.o -c /home/ddzzkk/PRAC/Test/cppcon2023/fifo5.cpp which is generated by this cmake command: cd build && cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE _EXPORT_COMPILE_COMMANDS=ON ../ the error is like this image.png (view on web) It says: ***@***.*** ~/P/T/c/build (main) [2]> /usr/bin/g++-12 -O3 -DNDEBUG -Wall -We rror -Wextra -Wconversion -Wno-unused-parameter -std=c++23 -o CMakeFiles/fifo5.dir/f ifo5.cpp.o -c /home/ddzzkk/PRAC/Test/cppcon2023/fifo5.cpp In file included from /usr/include/c++/12/atomic:41, from /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:3, from /home/ddzzkk/PRAC/Test/cppcon2023/fifo5.cpp:1: In member function ‘std::__atomic_base<_IntTp>::__int_type std::__atomic_base<_IntTp>::load(std::memory_order) const [with _ITp = long unsigned int]’, inlined from ‘auto Fifo5<T, Alloc>::pop() [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:207:49, inlined from ‘auto Fifo5<T, Alloc>::pop(T&) [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:218:18, inlined from ‘void Bench<T>::pop(value_type) [with T = Fifo5<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:90:42, inlined from ‘Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:52:20, inlined from ‘constexpr _Res std::__invoke_impl(__invoke_other, _Fn&&, _Args&& ...) [with _Res = void; _Fn = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:61:36, inlined from ‘constexpr typename std::__invoke_result<_Functor, _ArgTypes>::type std::__invoke(_Callable&&, _Args&& ...) [with _Callable = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:96:40, inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::_M_invoke(std::_Index_tuple<_Ind ...>) [with long unsigned int ..._Ind = {0}; _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:279:26, inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::operator()() [with _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:286:20, inlined from ‘void std::thread::_State_impl<_Callable>::_M_run() [with _Callable = std::thread::_Invoker<std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> > >]’ at /usr/include/c++/12/bits/std_thread.h:231:20: /usr/include/c++/12/bits/atomic_base.h:488:31: error: ‘long unsigned int __atomic_load_8(const volatile void*, int)’ writing 8 bytes into a region of size 0 overflows the destination [-Werror=stringop-overflow=] 488 | return __atomic_load_n(&_M_i, int(__m)); | ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~ In member function ‘std::__atomic_base<_IntTp>::__int_type std::__atomic_base<_IntTp>::load(std::memory_order) const [with _ITp = long unsigned int]’, inlined from ‘auto Fifo5<T, Alloc>::pop() [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:207:49, inlined from ‘auto Fifo5<T, Alloc>::pop(T&) [with T = long int; Alloc = std::allocator<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/Fifo5.hpp:218:18, inlined from ‘void Bench<T>::pop(value_type) [with T = Fifo5<long int>]’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:90:42, inlined from ‘Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>’ at /home/ddzzkk/PRAC/Test/cppcon2023/bench.hpp:57:20, inlined from ‘constexpr _Res std::__invoke_impl(__invoke_other, _Fn&&, _Args&& ...) [with _Res = void; _Fn = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:61:36, inlined from ‘constexpr typename std::__invoke_result<_Functor, _ArgTypes>::type std::__invoke(_Callable&&, _Args&& ...) [with _Callable = Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()>; _Args = {}]’ at /usr/include/c++/12/bits/invoke.h:96:40, inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::_M_invoke(std::_Index_tuple<_Ind ...>) [with long unsigned int ..._Ind = {0}; _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:279:26, inlined from ‘typename std::thread::_Invoker<_Tuple>::__result<_Tuple>::type std::thread::_Invoker<_Tuple>::operator()() [with _Tuple = std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> >]’ at /usr/include/c++/12/bits/std_thread.h:286:20, inlined from ‘void std::thread::_State_impl<_Callable>::_M_run() [with _Callable = std::thread::_Invoker<std::tuple<Bench<Fifo5<long int> >::operator()(long int, int, int)::<lambda()> > >]’ at /usr/include/c++/12/bits/std_thread.h:231:20: /usr/include/c++/12/bits/atomic_base.h:488:31: error: ‘long unsigned int __atomic_load_8(const volatile void*, int)’ writing 8 bytes into a region of size 0 overflows the destination [-Werror=stringop-overflow=] 488 | return __atomic_load_n(&_M_i, int(__m)); | ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~ cc1plus: all warnings being treated as errors I search for this error and don't know why it comes out. I debuged the code and found the problem seems to come from the usage of popper_t and its destructor, but I didn't figure out how to fix this. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>

[pffang]

Hi @FlameTornado10,

The performance issue may caused by the OS scheduler. Because WSL2 is a VM, sometimes the threads cannot get scheduled as expected.
The benchmark result may have a big difference with every try.

The compiler error is caused by you treating warning as error with parameter -We
rror.
The warning is an issue in GCC 12. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108374

[FlameTornado10]

@pffang
Thanks for your reply. I got the part about performance.
As for the error I encoutered, it seems like this warning is reported when the pointer could be nullptr, but in fifo5.hpp popCursor_ is simply a member variable rather than a pointer. So I'm still confused.