cuda-memcheck line numbering

I am trying to debug a memory access problem using cuda-memcheck. The out of bounds read provides an address and a source code line number, but that line number in the kernel is no where near the actual error. Are there some issues or restrictions concerning line numbering. These source files have a large number of #ifdef - #endif regions which are NOT compiled into the version I am trying to debug. Do those somehow affect the line numbering in memcheck. Runnign cuda 4.0 on linux redhat 5

1 Like