Currently, a caller is unable to determine the appropriate memory alignment that
is required for a given cipher. Thus, when using simple malloc, it may happen
that libgcrypt needs to add an additional copy step to re-align the input data.

If a caller would be able to determine the alignment on its own, the copy step
could be avoided and performance could be increased.

An example would be found in the Linux kernel crypto API where the
crypto_*_alignmask API calls allow the caller to know about alginment requirements.