my name is Rick and I am currently writing my bachelor's thesis in Computer Science using OpenCL 2.0's trying to accelerate network packet processing.
My system consists of an AMD A10-7890K Radeon R7 running Ubuntu 14.04.4 LTS in 64 bit and fglrx 15.20.3 [Sep 8 2015] driver version.
According to my implementation the system has Coarse Grained & Fine Grained SVM capabilities but not Fine Grained Atomics and Fine Grained System capabilities.
Question: Are those capabilities hardware or software dependant? Which system has Fine Grained System capabilities? Where can I find a list of CPUs and their SVM capabilities?
One of my ideas was to let the GPU produce network packets ready for sending directly into the transmit buffer of the network device, hence directly into kernel space. Therefore I came across Intel's article on SVM where they differentiate between multiple SVM capabilities OpenCL™ 2.0 Shared Virtual Memory Overview | Intel® Software.
Question: Is my idea of passing a kernel space address to an OpenCL kernel as a system SVM pointer even possible or did I misinterpret the meaning of system SVM?
Thanks for helping.