Thread: G++ PGO Blows Up

I've written a program that finds the order of the max clique in a graph. I'm trying to use profile guided optimizations (PGO) to squeeze a little extra performance out of it, b/c it's pretty maxed out in terms of algorithms. (That is, without using some weird math stuffs to further reduce the problem or optimize for the likely cases.)

I compile with the following makefile:

Code:

CXX        :=g++-4.8.1
OPTS    :=-O3 -flto -march=native -std=c++11 -pthread

all : first first-threaded first-pool second second-threaded \
    second-pool primes

debug : CXX        :=g++-4.8.1
debug : OPTS    :=-g -std=c++11 -pthread

gen-prof : OPTS    :=$(OPTS) -fprofile-generate=./pgo-data
use-prof : OPTS    :=$(OPTS) -fprofile-use=./pgo-data

gen-prof : all
    touch *.cpp

use-prof : all

debug : all

.PHONY : clean
clean :
    rm -f first first-threaded first-pool second second-threaded \
    second-pool primes *.o

first : main1.cpp graph.o util.cpp thread_pool.o
    $(CXX) $(OPTS) -o $@ $+

first-threaded : main1.cpp graph.o util.cpp thread_pool.o
    $(CXX) $(OPTS) -DTHREADS -o $@ $+

first-pool : main1.cpp graph.o util.cpp thread_pool.o
    $(CXX) $(OPTS) -DTHREADS -DPOOL -o $@ $+

second : main2.cpp graph.o util.cpp thread_pool.o
    $(CXX) $(OPTS) -o $@ $+

second-threaded : main2.cpp graph.o util.cpp thread_pool.o
    $(CXX) $(OPTS) -DTHREADS -o $@ $+

second-pool : main2.cpp graph.o util.cpp thread_pool.o
    $(CXX) $(OPTS) -DTHREADS -DPOOL -o $@ $+

primes : primes.cpp
    $(CXX) $(OPTS) -o primes primes.cpp

graph.o : graph.cpp
    $(CXX) $(OPTS) -c -o $@ $+

thread_pool.o : thread_pool/thread_pool.cpp
    $(CXX) $(OPTS) -c -o $@ $+

Then I run:
make gen-prof
./second 89 # leaves PGO data in ./pgo-data
make clean
make use-prof # errors, lots and lots of them

Here's some of the errors. In short, it's a lot of stuff about coverage mismatches and control flow of functions not matching their profile data.

Code:

g++-4.8.1 -O3 -flto -march=native -std=c++11 -pthread -fprofile-use=./pgo-data -o second main2.cpp graph.o util.cpp thread_pool.o
g++-4.8.1 -O3 -flto -march=native -std=c++11 -pthread -fprofile-use=./pgo-data -DTHREADS -o second-threaded main2.cpp graph.o util.cpp thread_pool.o
g++-4.8.1 -O3 -flto -march=native -std=c++11 -pthread -fprofile-use=./pgo-data -DTHREADS -DPOOL -o second-pool main2.cpp graph.o util.cpp thread_pool.o
main2.cpp: In function ‘thread_pool::async(std::function<void ()>&&)::{lambda()#2}::operator()() const [clone .isra.78]’:
main2.cpp:144:1: error: the control flow of function ‘_ZZN11thread_pool5asyncEOSt8functionIFvvEEENKUlvE0_clEv.isra.78’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
 }
 ^
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In function ‘__gnu_cxx::new_allocator<int>::allocate(unsigned long, void const*) [clone .isra.72]’:
main2.cpp:144:1: error: the control flow of function ‘_ZN9__gnu_cxx13new_allocatorIiE8allocateEmPKv.isra.72’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In function ‘std::allocator_traits<std::allocator<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void> > >::_S_destroy<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >(std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>&, std::allocator<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void> >::__destroy_helper*)std::enable_if<std::allocator<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void> >::__destroy_helper<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >::value, void>::type’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt16allocator_traitsISaINSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEEEE10_S_destroyISB_EENSt9enable_ifIXsrNSD_16__destroy_helperIT_EE5valueEvE4typeERSC_PSH_’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In function ‘std::get<0ul, std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1}>(std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1}&)std::__add_ref<std::tuple_element<0ul, std::tuple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1}> >::type>::type’:
main2.cpp:144:1: error: the control flow of function ‘_ZSt3getILm0EIZNSt13__future_base17_Async_state_implISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEC1EOSA_EUlvE_EENSt9__add_refINSt13tuple_elementIXT_ESt5tupleIIDpT0_EEE4typeEE4typeERSJ_’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)> >, (__gnu_cxx::_Lock_policy)2>::~_Sp_counted_ptr_inplace()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlS7_E_S6_EEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EED2Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::_M_get_deleter(std::type_info const&)’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base17_Async_state_implISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EE14_M_get_deleterERKSt9type_info’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::_M_dispose()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base17_Async_state_implISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EE10_M_disposeEv’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::~_Sp_counted_ptr_inplace()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base17_Async_state_implISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EED0Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::~_Sp_counted_ptr_inplace()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base17_Async_state_implISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EED2Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::_M_get_deleter(std::type_info const&)’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EE14_M_get_deleterERKSt9type_info’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::_M_destroy()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EE10_M_destroyEv’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::_M_dispose()’:
main2.cpp:144:1: warning: source locations for function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EE10_M_disposeEv’ have changed, the profile data may be out of date [enabled by default]
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>, std::allocator<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >, (__gnu_cxx::_Lock_policy)2>::~_Sp_counted_ptr_inplace()’:
main2.cpp:144:1: warning: source locations for function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEESaISB_ELN9__gnu_cxx12_Lock_policyE2EED0Ev’ have changed, the profile data may be out of date [enabled by default]
main2.cpp: In member function ‘std::thread::_Impl<std::_Bind_simple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1} ()> >::_M_run()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt6thread5_ImplISt12_Bind_simpleIFZNSt13__future_base17_Async_state_implIS1_IFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEC1EOSB_EUlvE_vEEE6_M_runEv’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::thread::_Impl<std::_Bind_simple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (&&)(std::function<void ()>))::{lambda()#1} ()> >, std::allocator<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (&&)(std::function<void ()>))::{lambda()#1} ()>, (__gnu_cxx::_Lock_policy)2>::_M_get_deleter(std::type_info const&)’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt6thread5_ImplISt12_Bind_simpleIFZNSt13__future_base17_Async_state_implIS2_IFZN11thread_pool5asyncEOSt8functionIFvvEEEUlS9_E_S8_EEvEC1EOSC_EUlvE_vEEEESaISI_ELN9__gnu_cxx12_Lock_policyE2EE14_M_get_deleterERKSt9type_info’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::thread::_Impl<std::_Bind_simple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (&&)(std::function<void ()>))::{lambda()#1} ()> >, std::allocator<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (&&)(std::function<void ()>))::{lambda()#1} ()>, (__gnu_cxx::_Lock_policy)2>::_M_destroy()’:
main2.cpp:144:1: warning: source locations for function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt6thread5_ImplISt12_Bind_simpleIFZNSt13__future_base17_Async_state_implIS2_IFZN11thread_pool5asyncEOSt8functionIFvvEEEUlS9_E_S8_EEvEC1EOSC_EUlvE_vEEEESaISI_ELN9__gnu_cxx12_Lock_policyE2EE10_M_destroyEv’ have changed, the profile data may be out of date [enabled by default]
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::thread::_Impl<std::_Bind_simple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (&&)(std::function<void ()>))::{lambda()#1} ()> >, std::allocator<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (&&)(std::function<void ()>))::{lambda()#1} ()>, (__gnu_cxx::_Lock_policy)2>::_M_dispose()’:
main2.cpp: In function ‘thread_pool::async(std::function<void ()>&&)::{lambda()#2}::operator()() const [clone .isra.78]’:
main2.cpp:144:1: warning: source locations for function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt6thread5_ImplISt12_Bind_simpleIFZNSt13__future_base17_Async_state_implIS2_IFZN11thread_pool5asyncEOSt8functionIFvvEEEUlS9_E_S8_EEvEC1EOSC_EUlvE_vEEEESaISI_ELN9__gnu_cxx12_Lock_policyE2EE10_M_disposeEv’ have changed, the profile data may be out of date [enabled by default]
main2.cpp:144:1: error: the control flow of function ‘_ZZN11thread_pool5asyncEOSt8functionIFvvEEENKUlvE0_clEv.isra.78’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
 }
 ^
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::thread::_Impl<std::_Bind_simple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1} ()> >, std::allocator<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1} ()>, (__gnu_cxx::_Lock_policy)2>::_M_get_deleter(std::type_info const&)’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt6thread5_ImplISt12_Bind_simpleIFZNSt13__future_base17_Async_state_implIS2_IFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEC1EOSC_EUlvE_vEEEESaISI_ELN9__gnu_cxx12_Lock_policyE2EE14_M_get_deleterERKSt9type_info’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp: In function ‘__gnu_cxx::new_allocator<int>::allocate(unsigned long, void const*) [clone .isra.72]’:
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp:144:1: error: the control flow of function ‘_ZN9__gnu_cxx13new_allocatorIiE8allocateEmPKv.isra.72’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::thread::_Impl<std::_Bind_simple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1} ()> >, std::allocator<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1} ()>, (__gnu_cxx::_Lock_policy)2>::_M_destroy()’:
main2.cpp:144:1: warning: source locations for function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt6thread5_ImplISt12_Bind_simpleIFZNSt13__future_base17_Async_state_implIS2_IFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEC1EOSC_EUlvE_vEEEESaISI_ELN9__gnu_cxx12_Lock_policyE2EE10_M_destroyEv’ have changed, the profile data may be out of date [enabled by default]
main2.cpp: In member function ‘std::_Sp_counted_ptr_inplace<std::thread::_Impl<std::_Bind_simple<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1} ()> >, std::allocator<std::__future_base::_Async_state_impl<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>::_Async_state_impl(thread_pool::async(std::function<void ()>&&)::{lambda()#2} (&&)())::{lambda()#1} ()>, (__gnu_cxx::_Lock_policy)2>::_M_dispose()’:
main2.cpp:144:1: warning: source locations for function ‘_ZNSt23_Sp_counted_ptr_inplaceINSt6thread5_ImplISt12_Bind_simpleIFZNSt13__future_base17_Async_state_implIS2_IFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEC1EOSC_EUlvE_vEEEESaISI_ELN9__gnu_cxx12_Lock_policyE2EE10_M_disposeEv’ have changed, the profile data may be out of date [enabled by default]
main2.cpp: In function ‘__static_initialization_and_destruction_0(int, int)’:
main2.cpp:144:1: error: the control flow of function ‘_Z41__static_initialization_and_destruction_0ii’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘__gnu_cxx::__concurrence_lock_error::~__concurrence_lock_error()’:
main2.cpp:144:1: error: the control flow of function ‘_ZN9__gnu_cxx24__concurrence_lock_errorD0Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘__gnu_cxx::__concurrence_unlock_error::~__concurrence_unlock_error()’:
main2.cpp:144:1: warning: source locations for function ‘_ZN9__gnu_cxx26__concurrence_unlock_errorD2Ev’ have changed, the profile data may be out of date [enabled by default]
main2.cpp: In member function ‘__gnu_cxx::__concurrence_broadcast_error::~__concurrence_broadcast_error()’:
main2.cpp:144:1: error: the control flow of function ‘_ZN9__gnu_cxx29__concurrence_broadcast_errorD0Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘__gnu_cxx::__concurrence_broadcast_error::~__concurrence_broadcast_error()’:
main2.cpp:144:1: error: the control flow of function ‘_ZN9__gnu_cxx29__concurrence_broadcast_errorD2Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘__gnu_cxx::__concurrence_wait_error::~__concurrence_wait_error()’:
main2.cpp:144:1: error: the control flow of function ‘_ZN9__gnu_cxx24__concurrence_wait_errorD0Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘__gnu_cxx::__concurrence_wait_error::~__concurrence_wait_error()’:
main2.cpp:144:1: error: the control flow of function ‘_ZN9__gnu_cxx24__concurrence_wait_errorD2Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::__future_base::_Result<void>::~_Result()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt13__future_base7_ResultIvED0Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In member function ‘std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda(std::function<void ()>&&)#1} (std::function<void ()>)>, void>::~_Deferred_state()’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlS6_E_S5_EEvED2Ev’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]
main2.cpp:144:1: note: use -Wno-error=coverage-mismatch to tolerate the mismatch but performance may drop if the function is hot
main2.cpp: In function ‘std::allocator_traits<std::allocator<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void> > >::_S_destroy<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >(std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void>&, std::allocator<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void> >::__destroy_helper*)std::enable_if<std::allocator<std::__future_base::_Deferred_state<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()>, void> >::__destroy_helper<std::_Bind_simple<thread_pool::async(std::function<void ()>&&)::{lambda()#2} ()> >::value, void>::type’:
main2.cpp:144:1: error: the control flow of function ‘_ZNSt16allocator_traitsISaINSt13__future_base15_Deferred_stateISt12_Bind_simpleIFZN11thread_pool5asyncEOSt8functionIFvvEEEUlvE0_vEEvEEEE10_S_destroyISB_EENSt9enable_ifIXsrNSD_16__destroy_helperIT_EE5valueEvE4typeERSC_PSH_’ does not match its profile data (counter ‘arcs’) [-Werror=coverage-mismatch]

All the errors seem to come from the second-threaded (and second-pool) compilation -- whatever changes in your code are happening when that THREADS symbol is defined, they are causing mismatches with the non-threaded version that you actually ran and profiled.

If there are some real differences there, you may want to set up your makefile to use different profile files for the different versions (rather than all using the same data file).

I should have noticed that second-pool was the only target failing. Thanks for pointing it out. I guess the massive wall of text confused me. (The errors I posted were only like 1/2 or 1/3.) I changed (gen,use)-prof to depend only on "second". When I used to same set of commands as before, I get an ICE (segfault) from gcc when I run use-prof. So I guess that might be a compiler error.

Code:

g++-4.8.1 -O3 -flto -march=native -std=c++11 -pthread -fprofile-use=./pgo-data -c -o graph.o graph.cpp
g++-4.8.1 -O3 -flto -march=native -std=c++11 -pthread -fprofile-use=./pgo-data -o second main2.cpp graph.o util.cpp thread_pool.o
main2.cpp: In function ‘chain_p(int)’:
main2.cpp:144:1: internal compiler error: Segmentation fault
 }
 ^
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://bugs.gentoo.org/> for instructions.
make: *** [second] Error 1