cublas<t>getrfBatched info in release mode

Accelerated Computing GPU-Accelerated Libraries
1

Does cublasgetrfBatched return valid info data in release mode? I am running this code:

status=cublasSgetrfBatched(cublasHandle,nlm_p3,d_pLs,nlm_p3,d_perm,d_perm+nlm_p3,1);
checkCudaErrors(cudaMemcpy(permutations, d_perm, perm_mem_size,cudaMemcpyDeviceToHost));
if (*(permutations+nlm_p3)!=0) 
	printf("*** warning: cublasSgetrfBatched info = %d (status=%d)***\n",*(permutations+nlm_p3),status);

In debug mode, info and status are both 0, and no warning appears. In release mode, I get info values on the order of +/- 1208978662, with status=0. My matrices are only 21 X 21. Is this normal? My end results appear ok… (I’m using MSVS 2012, & CUDA 5.5)

2

perm_mem_size should be equal to nlm_p3*sizeof(int). Is that the case?

3

Actually, I am transferring permutations and info in one array, adding the info to the nlm_p3+1 element. So, perm_mem_size = (nlm_p3+1)*sizeof(int);