It seems to me that cuda-memcheck doesn’t return as detailed information with CUDA Fortran as it does with CUDA C.
========= Invalid global read of size 8
========= at 0x00000408
========= by thread (0,31,8) in block (0,238)
========= Address 0x2015058f8 is out of bounds
Error: failed to read the error record (version mismatch?).
========= ERROR SUMMARY: 1 error
I compile the code using -Mcuda=cc20,cuda3.2 and mem-check is from cuda 3.2. Is there an alternative choice in CUDA Fortran?