nvcc 7.0 unable to compile with boost/atomic/detail/gcc-atomic.hpp included

[steinbac@…]

including boost/thread.hpp produces /usr/include/boost/atomic/detail/gcc-atomic.hpp(248): error: argument of type "const volatile uint8_t *" is incompatible with parameter of type "volatile void *"

$ nvcc -lboost_thread boost_thread.cu #include <boost/thread.hpp>

global void add_2(float a, float b, float *c){

*c = a+b;

}

int main(int argc, char *argv[]) {

float res = 0; float a = boost::thread::hardware_concurrency(); add_2<<<1,1>>>(a,52.,&res);

std::cout << boost::thread::hardware_concurrency() << "]" << res << "\n"; return 0;

}

system details:

fedora 21, 4.0.4-202.fc21.x86_64 gcc: gcc (GCC) 4.9.2 20150212 (Red Hat 4.9.2-6) boost: 1.55.0 Release 8.fc21 from system repos cuda 7.0.28

I already filed a bug report with nvidia

[squelsh@…]

Same here: Pretty annoying with Ubuntu 15.04. @steinbac: Could you please post a link to you Bugreport at NVidia? So I can track progress there?

[viboes]

Moving it to Boost.Atomic

[timblechmann]

could the reporter possibly double-check, if this still occurs with boost-1.58?

[anonymous]

Replying to timblechmann:

could the reporter possibly double-check, if this still occurs with boost-1.58?

The issue is indeed gone with boost/1.58.0 built with gcc 4.8.3. will check with gcc 4.9.2-6 by the end of the week.

[steinbac@…]

The problem disappears with cuda 7.5 as suggested by nvidia.

Nvidia bug information is:

---

Bug Information

---

Requester: Peter Steinbach

NVIDIA Bug ID: 1654452

Date: 06/11/2015 12:48 AM

Company/Division: CUDA RegDev Program [DevZone]

Severity: Medium Priority: 2 - Medium Synopsis: unable to compile with boost/atomic/detail/gcc-atomic.hpp included

Description: including boost/thread.hpp produces

/usr/include/boost/atomic/detail/gcc-atomic.hpp(248): error: argument of type "const volatile uint8_t *" is incompatible with parameter of type "volatile void *"

I double checked on a CentOS 7 box that comes with an "older" gcc (4.8.3) and the problem does not occur there with any cuda version!

[Andrey Semashev]

So, can we close this as the bug is fixed in nvcc?

[timblechmann]

closing as it is fixed in nvcc