Using CUDA, memory can be allocated on the GPU and shared to peers on
the PCIe bus such as the FPGA. Furthermore, the DMA on the GPU can also
be used to read and write to/from other memory on the PCIe bus, such as
BRAM on the FPGA.
Just using a standard std::list<> to hold plugins is problematic, because
we want to push Plugins to the list from within each Plugin's constructor
that is executed during static initialization. Since the order of static
initialization is undefined in C++, it may happen that a Plugin
constructor is executed before the list could be initialized. Therefore,
we use the Nifty Counter Idiom [1] to initialize the list ourself before
the first usage.
In short:
- allocate a buffer for the list
- initialize list before first usage
- (complicatedly) declaring a buffer is neccessary in order to avoid
that the constructor of the static list is executed again
