Illegal access errors only when compiling with -g -G parameters

Hi,

I’m new to cuda-gdb tool. I’m working on a remote server with 3 CUDA devices, 2 K80 and 1 GTX1060.

I tried to use cuda-gdb on a reliable code but when compiling with the -g -G flags the code returns the following GPU asserts:

GPUassert: an illegal memory access was encountered ****.cu 103
GPUassert: an illegal memory access was encountered ****.cu 104

Lines 103 and 104 contain the following instructions:

gpuErrchk( cudaMallocHost(&h_channel, sizeof(float) * f_size) );
gpuErrchk( cudaMemcpy(h_channel, d_channel, sizeof(float) * f_size, cudaMemcpyDeviceToHost) );

It worth noting that without the -g -G flags the code perfectly works.

Any idea?

Thank you in advance

Hi, CudaCrunch

This is cuda programming issue, suggest you post the question under https://devtalk.nvidia.com/default/board/57/cuda-programming-and-performance/

If possible, plesae provide the minor code that can reproduce the issue. Thanks!

Hi veraj,

Thank you for your answer. I’m not sure this is a cuda programming issue because the code perfectly works if not using cuda-gdb. I suspect the issue is related to how cuda-gdb handles multiple GPUs and X11.