Computation of the exponential integral is not supported by CUDA at this time. You may want to file an enhancement request to have this function added to CUDA through the bug reporting form linked from the registered developer website. If you do so, please prefix the subject line with “RFE:” to mark it as a “request for enhancement” rather than an actual bug. I will also be helpful to give some information on the use case for this function. Out of personal curiosity, what kind of application requires computation of the exponential integral?
For now you could either continue to use your current CPU-based implementation and ship the results to the GPU, or you could port CPU code to CUDA, e.g. by utilizing source code from the well-regarded Cephes library.