Does g++ with Mountain Lion supports -msse4.2? - c++

Question

I'm trying to run a program which uses __builtin_popcountll function.
When I compile the code using makefile which compiles source files with command/flags as shown below:
g++ -c -Wall `pkg-config opencv --cflags` -I./include -O3 -fopenmp -msse4.2 src/Utils.cpp -o src/Utils.o
It compiles without any error / warning. However, when I try to link the object (.o) files to build an executable, I get undefined symbols error.
Here is the command:
g++ src/BoostDesc.o src/Utils.o src/main.o `pkg-config opencv --libs` -lgomp -o main
and this is the complete error:
Undefined symbols for architecture x86_64:
"___builtin_popcountll", referenced from:
__ZN9boostDesc5Utils12matchHammingERKN2cv3MatERKSt6vectorIS2_SaIS2_EERS5_IS5_INS1_6DMatchESaISA_EESaISC_EE.omp_fn.0 in Utils.o
ld: symbol(s) not found for architecture x86_64
collect2: ld returned 1 exit status
make: *** [main] Error 1
I looked up the man page for gcc on Apple's website here, and it suggests that the flag works and I'm assuming it should work for g++ as well. Can someone confirm or refute the possibility of using this builtin function? Thnx!
g++ --version returns this:
i686-apple-darwin11-llvm-g++-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2336.11.00)

Answer 1

This optimization is specified either by doing: -mpopcnt or -march=corei7 to the compilation.
Take one sample:
% cat popcountl.c
int main(int argc , char** argv)
{
volatile long long x = 0xf0f0f0f0f0f0f0f0;
return __builtin_popcountll(x);
}
Turns out not to be supported in that version of g++:
~/D/e/popcountl [1]> /Applications/Xcode\ 4.6.3.app/Contents/Developer/usr/bin/g++ -mpopcnt -c popcountl.c
cc1plus: error: unrecognized command line option "-mpopcnt"
~/D/e/popcountl [1]> /Applications/Xcode\ 4.6.3.app/Contents/Developer/usr/bin/g++ -march=corei7 -c popcountl.c
popcountl.c:1: error: bad value (corei7) for -march= switch
popcountl.c:1: error: bad value (corei7) for -mtune= switch
~/D/e/popcountl> /Applications/Xcode\ 4.6.3.app/Contents/Developer/usr/bin/g++ --version
i686-apple-darwin11-llvm-g++-4.2 (GCC) 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2336.11.00)
Copyright (C) 2007 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
Works with Xcode5 g++ though.
% g++ --version
Configured with: --prefix=/Applications/Xcode.app/Contents/Developer/usr --with-gxx-include-dir=/usr/include/c++/4.2.1
Apple LLVM version 5.0 (clang-500.2.75) (based on LLVM 3.3svn)
Target: x86_64-apple-darwin12.5.0
Thread model: posix
g++ -mpopcnt -c popcountl.cpp
dhcp-3-127:~% otool -tV popcountl.o
popcountl.o:
:
0000000000000024 popcntq %rax, %rax
It looks like the flags for this optimization are not present in the 4.6 version of the XCode supplied compilers, but are present in the 5.0 version of the XCode compilers.
Bear in mind that you're actually running the llvm compiler with a g++ front-end, so it's not actually running official g++; this is probably the underlying reason for it not working properly in this case.