Cuda-memcheck error message not found on Google

Accelerated Computing CUDA CUDA Programming and Performance
1

I ran into this error while running cuda-memcheck. The error comes from a cudaMemsetAsync statement in a library that is too big to make an example here. The msesage is:

========= Host API memory access error at host access to 0x302000000 of size 13 bytes
=========     Invalid start allocation on access by cudaMemset destination.

where 13 is the number of bytes used in cudaMemsetAsync (just a number I used to test).

CUDA_CHECK(cudaMemsetAsync(data_ptr, c, 13, stream));

I tried to search the meaning of this error, but Google does not show any result. Does anyone know what could be the problem?

2

it might be better to ask this on the cuda-memcheck forum. Perhaps data_ptr is not properly allocated, or does not point to device memory allocation of length at least 13 bytes. It could also be a sign of host memory corruption (e.g. host stack corruption). It would also be useful to note whether the CUDA_CHECK macro reports anything for this call when run under cuda-memcheck.