Does cublas support 64 bit integers (long longs)
I do not think so, but doubles (64 bit floating point) and complex double are supported.
Current CUBLAS documentation can be found here: http://docs.nvidia.com/cuda/cublas/index.html
I guess the question may be with regard to the dimension arguments, rather than matrix and vector element type?
The current version of CUBLAS does not support ‘long long int’, either as an element type, or to specify dimensions.
Like the reference BLAS, CUBLAS supports elements of type float (functions starting with ‘S’), double (functions starting with ‘D’), single-precision complex (functions starting with ‘C’) and double-precision complex (functions starting with ‘Z’).
Dimensions arguments use ‘int’ like the reference BLAS, so are limited to [1,2**31-1]. This should be sufficient for current GPU memory capacities, as a maximum size vector of float (the smallest data type supported) would comprise 8 GB of data.