--- Quote Start ---
This is bad news. I want use "autorun" kernel with global memory opened to host for omit Altera/Intel hidden/secret interface to data exchange and make my "signals" and "replies" back in memory.
In aocl_programming_guide.pdf (2.3.1.2) is no words about "no interface to memory", only 0 with host.
Or Altera interface from host with OpenCL kernels is open ? Where I may read about it ?
We want cut a working OpenCL core in VHDL and insert it to big FPGA project with more other no-OpenCL blocks, it is possible ?
--- Quote End ---
You cannot allocate global memory from kernel, it has to be allocated on the host and the pointer passed onto the kernel. Since autorun kernels cannot have an interface to the host, they cannot receive the pointer to global memory and hence, direct communication between autorun kernels and external memory is impossible. Basically, an autorun kernel cannot have any arguments and it is not at all visible from the host. The only means of communication with an autorun kernel is through channels.
On the subject of connecting an OpenCL kernel to a custom memory interface, that is not going to work unless your interface is 100% compatible with Altera's interface. I am not really sure whether their interface is open or not, but my guess is that it is not. For your specific usage, I think the new Intel HLS tool would be a much better choice since it allows you to convert a C/C++ component to an HDL module with a standard Avalon interface (Memory-mapped, streaming, etc.) and can be connected to any other HDL module using Qsys. You also have the choice of integrating your existing HDL modules inside of an OpenCL project, but the modules would need to comply with Altera's interface and communication with DDR/PCI-E/Network will have to go through Altera's interface.
Honored Contributor
