Member Declarations

Let us start with declaring the new member-functions in every necessary header, assuming that you want both a CPU and a GPU routine to be implemented. In the X_default_cpu.h and X_default_gpu.h header-files, you will find function-declarations of the form void routine() in the public section and void routine_() (note the underscore) in the private section of the class interface. This is where the new routine should be declared in two-fold: with and without a trailing underscore. The non-underscored version is merely a wrapper of the underscored version, and takes care of the timing (if necessary). Each of the non-underscored routines has an inline implementation in the same file, which looks like this (same for DefaultGPU):

Assuming there are already other inline implementations present, you can simply copy-paste and change the routine-identifier.

The Hybrid_ class also contains function-declarations in its public section (non-underscored). The implementation is again a wrapper, this time of either the CPU or GPU version. Exactly which one is determined by the DevicePolicies, which we will come to later. Their inline implementation (in the same file) should look like this:

This is also the tricky part. You have to figure out for yourself which device has the data that is used by this particular routine (SourceDevice), and make sure you synchronize with it.