Hello,
I’m using Cubals with CUDA 4.0 on Tesla M2070 GPU.
I got q segfault every time I passe a scalar by ref to a Cublas kernel
...
cublasHandle_t handle;
cudaMalloc((void**)&dot_GPU, sizeof(double));
cudaMemset(dot_GPU, 0, sizeof(double));
cudaStreamCreate(&stream_cublas);
cublasCreate(&handle);
cublasSetStream(handle, stream_cublas);
/* This one is OK */
cublasDdot(handle,n,_vect,1,_vect,1,&dot_RAM);
/* And this one is KO :) */
cublasDdot(handle,n,_vect,1,_vect,1,dot_GPU);
This is a example of a dot kernel, by i get the same thing with scal and axpy.
Any ideas??
Stole