Segfault with CUDA-2.3, gcc-4.3 on ubuntu(x86_64)

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

the following simple program gives me a segmentation fault

__global__ void foo(float a, float b, float c) {}

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

{

	int i;

	for (i=0; i<11; i++)

		foo<<<4, 256>>>(1.0, 2.0, 3.0);

	return 0;

}

when I compile it with the following command:

nvcc --compiler-bindir=/usr/bin/gcc-4.3 --device-emulation -g -O3 -o test test.cu

Some more info on my environment:

(uname) Linux 2.6.31-19-generic #56-Ubuntu SMP x86_64 GNU/Linux

(nvcc --version) Cuda compilation tools, release 2.3, V0.2.1221

Note I’m using gcc-4.3 due to problems with CUDA on gcc-4.4 and I’m using the device emulation mode. Also the segfault only occurs when compiling with -O3.

Excerpts from valgrind output (not sure if this helps):

==5022== Conditional jump or move depends on uninitialised value(s)

==5022==	at 0x4C271E3: memcpy (mc_replace_strmem.c:77)

==5022==	by 0x4E49D7E: (within /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x4E462BE: cudaLaunch (in /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x400C15: main (cuda_runtime.h:713)

==5022==

==5022== Conditional jump or move depends on uninitialised value(s)

==5022==	at 0x4C271F8: memcpy (mc_replace_strmem.c:402)

==5022==	by 0x4E49D7E: (within /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x4E462BE: cudaLaunch (in /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x400C15: main (cuda_runtime.h:713)

==5022==

==5022== Conditional jump or move depends on uninitialised value(s)

==5022==	at 0x4C27245: memcpy (mc_replace_strmem.c:402)

==5022==	by 0x4E49D7E: (within /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x4E462BE: cudaLaunch (in /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x400C15: main (cuda_runtime.h:713)

==5022==

==5022== Use of uninitialised value of size 8

==5022==	at 0x4C27264: memcpy (mc_replace_strmem.c:402)

==5022==	by 0x4E49D7E: (within /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x4E462BE: cudaLaunch (in /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x400C15: main (cuda_runtime.h:713)

==5022==

==5022== Process terminating with default action of signal 11 (SIGSEGV)

==5022==  Bad permissions for mapped region at address 0x65B1D2E

==5022==	at 0x4C27264: memcpy (mc_replace_strmem.c:402)

==5022==	by 0x4E49D7E: (within /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x4E462BE: cudaLaunch (in /home/graviton/software/cuda/lib64/libcudart.so.2.3)

==5022==	by 0x400C15: main (cuda_runtime.h:713)

Thanks,

graviton