cuda-gdb debugger stalls


i have a Cuda application that performs some math calculations on a molecular modelling simulation.
I use openGl coded in C++ so there is no graphics interop code in Cuda

I have tried on a Fermi card (gt 430) on my local pc and the debugger crashes when calling the kernell. It passes the cudaMalloc calls
compile flags : -gencode arch=compute_20,code=sm_20

I have tried it remotely on a Tesla c2070 which is not used for X11. gdb crashes on the first cudaMalloc call.
compile flags : -gencode arch=compute_20,code=sm_20
-gencode arch=compute_10,code=sm_10 : fails the compilation on atomics so could not be used …

Any clues…?


have you try to compile it with “-g -G”? Have you try cuda-memcheck?