No way to write output to device memory.

[simdeistud]

The gpujpeg_writer_init function allocates memory on the host with malloc, without giving the option to the user to instead allocate on the GPU memory or with pinned memory. If someone needs to keep processing the image on the GPU, a lot of time is wasted for host-device transfers. If this can be done in some other way or if there are reasons why this is impossible, it would be useful to document it, as other libraries make the device buffer available to the user.