Why does caldgemm give such an error? CUDA Error 77: an illegal memory access was encountered

[Codingrunningman]

I compile it successfully and I use cuda as the backend. When I execute this command './dgemm_bench -O 2 -p -A -B -m 40960 -n 40960', it always poses such an error: CUDA Error 77: an illegal memory access was encountered.

Use -? for help
Initializing CALDGEMM (CUDA Runtime)
Cannot use multiple devices without multithreading
Running on 1 devices with 30 bbuffers (gn01)
Initializing Data... ...alloc A (330240 KB) B (327744 KB) C (13109760 KB)......init A...init B...Done
Doing initial run... WARNING, you are using GPU_C or '-o g' option, but apparently you did not use CALDGEMM memory allocation with gpu_accessible feature ('-_' is missing).
You must take care to allocate GPU accessible memory yourself, or this can lead to invalid memory accesses.
CUDA Error 77: an illegal memory access was encountered
caldgemm_cuda.cu:371
CUDA Kernel Execution
Error running CALDGEMM
exiting

How can I solve this problem? And I can not change the backend because my gpu belongs to nvidia, right?

[davidrohr]

From, I cannot tell what is the problem except that apparently the GPU kernel crashed. Please note that caldgemm is not developed further and I cannot give support any more.

[Codingrunningman]

Thanks